Текст
                    The Principles of
Nonlinear Optics
Y. R. SHEN
University of California, Berkeley
A Wiley-Interscience Publication
JOHN WILEY & SONS
New York Chichester Brisbane Toronto Singapore


Preface The laser is certainly one of the greatest inventions in the history of science. Its arrival, a quarter of a century ago, created many fascinating new fields, among which nonlinear optics undoubtedly has the broadest scope and the most influential proponents. The field originated with the experimental work of P. A. Franken and co-workers on optical second-harmonic generation in 1961 and the theoretical work of N. Bloembergen and co-workers on optical wave mixing in 1962. Since that time the field has grown at such a prodigious rate that today it has already found applications in nearly all areas of science. The very broad expanse of nonlinear optics certainly is most exciting, but it also makes the field difficult to comprehend. The vast amount of knowledge generated over the years is scattered everywhere in the literature. Beginners in nonlinear optics often have a hard time acquainting themselves with the many facets of the field. Even workers in nonlinear optics may sometimes have difficulty finding some rudimentary information about a subarea of the field they are not familiar with. A book on nonlinear optics offering a fair introduc- introduction to all branches of the field is clearly needed. Actually, there already exist a number of books on the subject of nonlinear optics. The most authoritative, by N. Bloembergen, lays the foundation for nonlinear optics. However, since the book was written in 1965, it is clearly outdated, as is the 1964 book by S. A. Akhmanov and R. V. Khokhlov (English translation 1972). Among the remaining books found in most academic libraries, some are elementary or narrow in scope and others tend to con- concentrate on special topics of nonlinear optics. Conference proceedings may provide a broader perspective but usually are very advanced and lack continu- continuity. What one would like to have is a book that not only logically presents ihe basic principles of nonlinear optics but also systematically describes the subareas of the field. This book is meant to fulfill that need. To write a book covering the entire field of nonlinear optics in depth is an impossible task for a single author. This book falls short of the full details in the description of some subject matter. Furthermore, to limit the size of the book several areas were omitted. These include collision-induced nonlinear optical excitations, optical multistabilities, bifurcations and chaos, quantum statistics of nonlinear optics, and many highly nonlinear optical effects. In writing the book, I chose to emphasize the fundamentals as well as the interplay between theory and experiment. Physical concepts are stressed in the
theoretical presentation, although equations are usually unavoidable for a careful account. In the illustration of a particular process, a brief description of the experimental situation is given to provide the readers with a realistic picture. References at the end of each chapter supplement the omitted details in the text, but each list is purposely brief. This book grows out of a graduate physics course on modern optics I taught several times at Berkeley. The frustrating experience of selecting appropriate materials for the course led me to write this book. It was therefore prepared at a level intended for a physics graduate student. With some effort, students of chemistry and engineering who are serious in learning about nonlinear optics should also be able to appreciate the material without much difficulty. And the book should be a useful reference for professionals in their review of the field. The book begins with a general introduction, followed by a description of the fundamentals in Chapters 2 and 3. Electrooptical and magnetooptical effects are considered in Chapter 4 as special nonlinear optical phenomena, and their inverse effects are discussed in Chapter 5. The more familiar second-order nonlinear optical effects are discussed in Chapters 6 to 9, and the third-order effects are considered in Chapters 10 to 17. In the discussion, parametric conversion in Chapter 9 is treated as the inverse of a mixing process. Stimulated light scattering is shown in Chapters 10 to 11 to behave like a parametric process from the general coupled-wave point of view, although it is often understood as a two-photon process resulting in material excitation. While the first half of the book deals with the traditional type of nonlinear optics, the second half is on special topics. Chapters 13, 15, and 18 to 28 are devoted to the discussion of various nonlinear optical effects and applications which have fascinated researchers in recent years. In many of these areas, new results and discoveries are still being reported frequently at meetings and in journals. Some parts of the text are bound to become obsolete sooner or later, but it is hoped that the principles should always remain unchanged. As a tribute to the twenty-fifth anniversary of the invention of lasers, this book is written to manifest a part of the intellectual wealth lasers have created. I am deeply indebted to Professor Bloembergen for introducing nonlinear optics to me during its early development stage. His teaching and guidance have led to the great satisfaction I have experienced in research in the past 20 years. I should like to express my gratitude to all my friends and colleagues who have supported my effort in writing this book. Special thanks are due to S. J. Gu, whose critical reading of the manuscript resulted in many changes and corrections. I also thank T. F. Heinz, X. D. Zhu, M. Mate, Y. Twu, and many others for their contribution in proofreading and improving the manuscript. With respect to the preparation of the manuscript, I, am most grateful to Rita Jones, who not only typed the entire manuscript but also helped and supported the project in all possible ways. Without her devoted effort, completion of this book would not have been possible. Finally, my wife,
Hsiao-Lin, deserves my warmest appreciation. It is her patience, understand- understanding, encouragement, and help in many details that gave me faith and strength in the course of writing this book. Y. R. Shen flertefry. California April 1984
Contents 1 Introduction 1 2 Nonlinear Optical Susceptibilities 13 3 General Description of Wave Propagation in Nonlinear Media 42 4 Electrooptical and Magnetooptical Effects 53 5 Optical Rectification and Optical Field-Induced Magnetization 57 6 Sum-Frequency Generation 67 7 Harmonic Generation 86 8 Difference-Frequency Generation 108 9 Parametric Amplification and Oscillation 117 10 Stimulated Raman Scattering 141 Π Stimulated Light Scattering 187 12 Two-Photon Absorption 202 13 High-Resolution Nonlinear Optical Spectroscopy 211 14 Four-Wave Mixing 242 15 Four-Wave Mixing Spectroscopy 266 16 Optical-Field-Induced Birefringence 286 17 Self-Focusing 303 18 Multiphoton Spectroscopy 334
χϊϊ Contents 19 Detection of Rare Atoms and Molecules 349 20 Laser Manipulation of Particles 366 21 Transient Coherent Optical Effects 379 22 Strong Interaction of light with Atoms 413 23 Infrared Multiphoton Excitation and Dissociation of Molecules 437 24 Laser Isotope Separation 466 25 Surface Nonlinear Optics 479 26 Nonlinear Optics in Optical Waveguides 505 27 Optical Breakdown 528 28 Nonlinear Optical Effects in Plasmas 541 Index 555
1 Introduction Physics would be dull and life most unfulfilling if all physical phenomena around us were linear. Fortunately, we are living in a nonlinear world. While linearization beautifies physics, nonlinearity provides excitement in physics. This book is devoted to the study of nonlinear electromagnetic phenomena in the optical region which normally occur with high-intensity laser beams. Nonlinear effects in electricity and magnetism have been known since Maxwell's time. Saturation of magnetization in a ferromagnet, electrical gas discharge, rectification of radio waves, and electrical characteristics of p-n junctions are just a few of the familiar examples. In the optical region, however, nonlinear optics became a subject of great common interest only after the laser was invented. It has since contributed a great deal to the rejuvenation of the old science of optics. 1.1 HISTORICAL BACKGROUND The second harmonic generation experiment of Franken et al.1 marked the birth of the field of nonlinear optics. They propagated a ruby laser beam at 6942 A through a quartz crystal and observed ultraviolet radiation from the crystal at 3471 Ä. Franken's idea was simple. Harmonic generation of electro- electromagnetic waves at low frequencies had been known for a long time. Harmonic generation of optical waves follows the same principle and should also be observable. Yet an ordinary light source is much too weak for such an . experiment. It generally takes a field of about 1 kV/cm to induce a nonlinear response in a medium. This corresponds to a beam intensity of about 2.5 kW/cm2. A laser beam is therefore needed in the observation of optical harmonic generation. Second harmonic generation is the first nonlinear optical effect ever ob- observed in which a coherent input generates a coherent output. But nonlinear optics covers a much broader scope. It deals in general with nonlinear
interaction of light with matter and includes such problems as light-induced changes of the optical properties of a medium. Second harmonic generation is then not the first nonlinear optical effect ever observed. Optical pumping is certainly a nonlinear optical phenomenon well known before the advent of lasers. The resonant excitation of optical pumping induces a redistribution of populations and changes the properties of the medium. Because of resonant enhancement, even a weak light is sufficient to perturb the material system strongly to make the effect easily detectable. Low-power CW atomic lamps were used in the earlier optical pumping experiments on atomic systems. Optical pumping is also one of the effective schemes for creating an inverted population in a laser system. In general, however, observation of nonlinear optical effects requires the application of lasers. Numerous nonlinear optical phenomena have been discovered since 1961. They have not only greatly enhanced our knowledge about interaction of light with matter, but also created a revolutionary change in optics technology. Each nonlinear optical process may consist of two parts. The intense light first induces a nonlinear response in a medium, and then the medium in reacting modifies the optical fields in a nonlinear way. The former is governed by the constitutive equations, and the latter by the Maxwell's equations. At this point, one may raise a question: Are all media basically nonlinear? The answer is yes. Even in the case of a vacuum, photons can interact through vacuum polarization. The nonlinearity is, however, so small that with currently available light sources, photon-photon scattering and other nonlinear effects in vacuum are still difficult to observe.2 So, in a practical sense, a vacuum can be regarded as linear. In the presence of a medium, the nonlinearity is greatly enhanced through interaction of light with matter. Photons can now interact much more effectively through polarization of the medium. 1.2 MAXWELL'S EQUATIONS IN NONLINEAR MEDIA All electromagnetic phenomena are governed by the Maxwell's equations for the electric and magnetic fields E<r, f) and B(r, r): 1 dB νχΕ=-71Γ· V Χ Β = —τ— + —— J, (ι ι! c dt c U-l) V ■ Ε = 4πρ, V -B = 0 where J(r, /) and p(r, t) are the current and charge densities, respectively. They
Maxwell's Equations in Nonlinear Media are related by the charge conservation law We can often expand J and ρ into series of multipoles:3 A.2) -^(V ■ Q) = p0- V P- V(V Here, P, M, Q,..., are respectively the electric polarization, the magnetization, the electric quadrupole polarization, and so on. However, as pointed out by Landau and Lifshitz,4 it is not really meaningful in the optical region to express J and ρ in terms of multipoles because the usual definitions of multipoles are unphysical. In many cases, for example in metals and semicon- semiconductors, it is more convenient to use J and ρ directly as the source terms in the Maxwell's equations, or to use a generalized electric polarization Ρ defined by where Jdc is the dc current density- In other cases, the magnetic dipole and higher-order multipoles can be neglected. Then, the generalized Ρ reduces to the electric-dipole polarization P. The difference between Ρ and Ρ is that Ρ is a nonlocal function of the field and Ρ is local. In this book, we assume electric dipole approximation, Ρ = Ρ, unless specified. -*■ With A.2) and A.4), the Maxwell's equations appear in the form 1 dB V Χ Ε- 3", e at V -(E + 4ttP) = 0, V -B = 0 where Ρ is now the only time-varying source term. In general, Ρ is a function of Ε that describes fully the response of the medium to the field, and it is often known as the constitutive equation. If we could just write the constitutive equation and find the solution for the resulting set of Maxwell's equations with appropriate boundary conditions, then all optical phenomena would be pre- predictable and easily understood. Unfortunately, this seldom is possible. Physi- Physically reasonable approximations must be resoried to in order to make the mathematical solution of the equations feasible. This is where physics comes into play.
The polarization Ρ is usually a complicated nonlinear function of E. In the linear case, however, Ρ takes a simple linearized form P(r,0=/°° XA)(r-r'.'-'')-E(r\ t')dr'dt' A.6) where χ is the linear susceptibility. If Ε is a monochromatic plane wave with E(r, f) = E(k, ω) = ^(k, w)exp(ik-r — ϊωί), then Fourier transformation of A.6) yields the familiar relation P(x,t)-P(k,u) )E(k) with XA)(k,«)= f" xA)(r, (Jexpi-ikT + iw/lrfnif. A.8) The linear dielectric constant e(k, ω) is related to xA)(k, ω) by e(k,«) = l + 4irxA>(k,w). A.9) In the electric dipole approximation, x(IJ(r, r) is independent of r, and hence both xA)(k, ω) and e(k,«) are independent of k. In the nonlinear case, when £ is sufficiently weak, the polarization Ρ as a function of Ε can be expanded into a power series of E: P(r, () = /" XAV -r',f- f')-E(r,f')rfr'i* + /_" Xm(r - η, t - v.w -rltt~ ij):^, (J χΕ^,ί^^Λ^Γ,Λ1 where χ(π) is the nth-order nonlinear susceptibility. If Ε can be expressed as a group of monochromatic plane waves
Anharmonic Oscillator Model 5 then, as in (he linear case, Fourier transform of A.10) gives P(k, ω) - P<»(k, ω) + PB)(k. ») + PC'(k, «) + · · ■ A.12) with PB|(k, ω) - x<2>(k - k,- + k}, u = ω,- + «,): E(k„ «,)E(kr «,), PE>(k, ω) = x<3)(k = k,- + k, + k,, u = ω,- + ω, + ω,) and X<n)(k = k, + k2 + ■ ■ ■ + k„, u = c^ + w2 + ■■■«„) Again, in the electric dipole approximation, x(n)(r, () is independent of r, or X(ll)(k, ω) independent of k. The linear and nonlinear susceptibilities characterize the optical properties of a medium. If χ(β) is known for a given medium, then at least in principle, the nth-order nonlinear optical effects in the medium can be predicted from the Maxwell's equations in A.5). Physically, χ<π) is related to the microscopic structure of the medium and can be properly evaluated only with a full quantum-mechanical calculation. Simple models are, however, often used to illustrate the origin of optical nonlinearity and some characteristic features of χ1. We consider here the anharmonic oscillator model and the free electron gas model. U ANHARMONIC OSCILLATOR MODEL In this model, a medium is composed of a set of Ν classical anharmonic oscillators per unit volume. The oscillator describes physically an electron bound to a core or an infrared-active molecular vibration. Its equation of motion in the presence of a driving force is A.15)
6 Introduction We consider here the response of the oscillator to an applied field with Fourier components at frequencies ±ωι and ±ω;: F = ^ [E,(*--.' + «'">') + £,(«"'-.' + e-")I· A.16) The anharmonic term ax2 in A.15) is assumed to be small so that it can be treated as a perturbation in the successive approximation of finding a solution: The induced electric polarization is simply P = Nqx. A.18) The first-order solution is obtained from the linearized equation of A.15): ) + c.c. where c.c. is a complex conjugate. Then, the second-order solution is obtained from A.15) by approximating ax7 by ax{1I: (, 2) (,) B) + x«>@) + c.c (ω ± ω ) = -2aDi/mJ£1£2 (Ug - Ιύ\ - iW,r)(tJo - td; Τ '«21") VI-'(«l±"i)' "o -(«ι ± ) - i("i ± "j)rj A.20) (ω2 - ω,2 - ίω,ΓJ(<4 - 4ω,2 - /2«,Γ) / ί \2 1 / 1 . 1 By successive iteration, higher-order solutions can also be obtained. As seen in the second-order solution, new frequency components of the polarization at <■>! ± ω3, 2«ι, 2ω2, and 0 have appeared through quadratic interaction of the field with the oscillator via the anharmonic term. The oscillating polarization components will radiate and generate new em waves at «ι ± «2> 2«i, and 2ω2-
Anharmonic Oscillator Mode] 7 Thus, sum- and difference-frequency generation and second-harmonic genera- generation are readily explained. The appearance of the zero-frequency polarization component is known as optical rectification. More generally, frequency compo- components at ω = η,«! ± η2ω2, with π, and n2 being integers, are expected in the higher-order solutions. In this model, the anharmonicity a determines the strength of the nonlinear interaction. Treating ax2 as a small perturbation in the foregoing calculation is equiva- equivalent to the assumption that Ε is small and Ρ can be expanded into power series of E. We can give a rough estimate on how the nonlinear polarization should diminish with increasing order. Assuming the nonresonani case with ω0 » ωι and ω2, we find from A.19) and A.20), ρω qaE A.21) For an electron bound to a core, if χ is so large that the harmonic force muijj and the anharmonic force max2 are of the same order of magnitude, then both will be of the same order of magnitudes as the total binding force on the electron \qEal\: \qEa,\ - mu^x ~ max7 A.22) Equation A.21) then becomes ρφ In fact, one can show in general A.23) A.24) such that \E/Ea!\ acts as an expansion parameter in the perturbation calcula- calculation. Typically, Εαι~3χ 108 V/cm. The Ε field for a 2.5-W/W laser beam is only 30 V/cm with \E/Ea,\ ~ 10"'. The nonlinear polarization is much weaker than the linear polarization. This suggests that the observation of nonlinear optical effects requires high-in tensity laser beams. Relation A.24), however, is true only for optical frequencies away from resonance. Near resonance, the resonant denominators may drastically en-
hance the ratio |/>ι"+1)/Ρ(''Ί- Consequently, the nonlinear effects can be detected with much weaker light intensity. Optical pumping is an example. With resonant enhancement, it may even happen that \Pin+ V/pW\ > 1. When this is the case, the perturbation expansion is no longer valid, and the full nonlinear expression of Ρ as a function of Ε must be included in the calculation. The problem then falls into the domain of strong interaction of light with matter. 1.4 FREE ELECTRON GAS A simple but realistic model to illustrate optical nonlinearity in a medium is the free electron gas model. It properly describes the optical properties of an electron plasma. The simplified version of the model starts with the equation of motion for an electron A.25) Damping is neglected here for simplicity. Clearly, the only nonlinear term in this equation is the Lorentz force term. Since υ ■* c in a plasma, the Lorentz force is much weaker than the Coulomb force, and then (e/mc)v Χ Β in A.25) can be treated as a perturbation in the successive approximation of the solution. For Ε = £ie!k<'-'w*' + (f2elk''r"'"i'' + c.c, we obtain , x(k2 X i2) + «?2 x(k, Χ <f,) and so on. For a uniform plasma with an electronic charge density p, the current density is given by with, for example,
Free Electron Gas 9 and so on. This shows explicitly how an electron gas can respond nonlinearly to the incoming light through the Lorentz term. In a more rigorous treatment of an electron gas, we must also lake into account the spatial variations of the electron density ρ and velocity v. Two equations, the equation of motion and the continuity equation,5 are now necessary to describe the electron plasma: and A.28) ^+ V(pv) = 0 where ρ is the pressure and m is the electron mass. The pressure gradient term in the equation of motion is responsible for the dispersion of plasma reso- resonance, but in the following calculation we assume vp = 0 for simplicity. Then, coupled with A.28), is the set of Maxwell's equations 1 dB c dt v XB_I^ = i^ = i5PI A.29) c dt c c ' VE = 4w{p-p(m), and VB-O We assume here that there is a fixed positive charge background in the plasma to assure charge neutrality in the absence of external perturbation. Successive approximation can be used to find J as a function of Ε from A.28) and A.29). Let6 ρ - ρ«) + ρθ> + pW + . . . , V = T0) + f (D + . . . , and j = j(i) + jP) + ... A.30) with j(') = pCV» and A.31)
Introduction We shall find the expression for jB)Btd) as an example assuming E exp(ik-r - ίωί). Substitution of A.30) into A.28) and A.29) yields = -ί,,,ν«» --^1 A-32) M_ -/2ων<2)= - »-— ϊA»ΧΒ. me The second-order current density is then given by ip<°>\ e2 Γ(Ε- 2ω [mV m'wc + ·4 " (? E)E. The last term in A.33) has the following equalities: A.33) Vp|0> ■ Ε 1 - ωl A.34) where ωρ = Dwp(O1e/m)i/! is the plasma resonance frequency. With A.34), Β = (e/w)v X E, and the vector relation Ε X (V Χ Ε) + (Ε- ν)Ε = i V(E · Ε), the current density in A.33) can be written as Am V 4m V V(E-E) ie2 | ypl0>-E 1 - «>2 E. A.35) Equation A.33) shows exph'citly that aside from the Lorentz term, there are also terms related to the spatial variation of E. They actually arise from the nonuniformity of the plasma. In a uniform plasma, vp@) = 0 and hence VE = 0 from A.34). This means that k is perpendicular to Ε and therefore (Ε · ν )E also vanishes. The Lorentz term is then the only term in J|2lBu).
The induced current density JB)Bu) should now act as a source for second harmonic generation in the plasma. In a uniform plasma with a single pump beam, J<2)B«) α Ε χ Β is only along the direction of beam propagation. Since an oscillating current cannot radiate longitudinally, no coherent second harmonic generation along the axis of beam propagation is expected from the bulk of a uniform plasma. In the bulk of a nonuniform plasma or at the boundary surface of a uniform plasma, however, it is possible to find second harmonic generation through the nonvanishing Vp@). Equation A.35) shows that when vp@> Ψ 0, the nonlinear response of the medium JB>B(d) is greatly enhanced if ω is near the plasma resonance. From the general principle, the nonlinear response of a medium is resonantly enhanced when the incoming field hits the resonance of the medium. One should, of course, also expect a resonant enhancement in the second harmonic generation when 2ω is at resonance. This actually does come in through the response of the second harmonic field to JB)Bu). As 2ω -· ωρ, the current density will excite the longitudinal field at 2ω resonantly. As shown in A.33) or A.35), the current density JaBio) depends exclu- exclusively on the spatial variation of E. In fact, using vector identities, the expression of J<2)Bw) in A.33) or A.35) can be put into the form7 JB)Bij) = cV X ( ) - fwv ■ ( ). Comparing it with A.3), we recognize that the two terms in J<2)B«) represent the magnetic dipole and electric quadrupole contributions, respectively. No induced electric dipole polarization exists in a plasma. Also, the induced electric quadrupole polarization depends on the gradient of the electric field, and therefore cannot show up in the bulk of a uniform plasma. The free electron model here is applicable to a number of real problems. First, it can be used to describe the optical nonlinearities due to plasmas in metals and semiconductors. Second harmonic generation from metal surfaces is readily observable,6 Then, with some modification to take into account the net charge distribution, the nonvanishing v/>, and so on, it can also be used to describe the optical nonlinearities of a gas plasma. Various nonlinear optical effects in gas plasmas have been observed. They will be discussed in some detail in Chapter 28. The model has also been used to describe the observation of nonlinear effect in a crystal in the X-ray region.3 The electron binding energy is much weaker than the X-ray photon energy, and therefore the electrons in the crystal will respond to the X-ray as if they were free. REFERENCES 1 P. A. Franken, A. E. Hill, C. W. Peters, and G. Weinreich, Phys. Rev. Uli. 7, 118 A961). 2 See, for example, G. Rosen and F. C. WhiUnore, Phys. Rev. 1J7B, 1357 A965). 3 J. D. Jackson, Classical Electrodynamics (McGraw-Hill, New York 1975), 2nd ed., p. 739; W. K. H. Panofsky and M. Phillips, Classical Electricity and Magnetism (Addison-Wesley, Reading, Mass., 1962), p. 131.
12 Introduction 4 L. D. Landau and Ε. Μ. Lifshitz, Electrodynamics in Continuous Media (Pergamon Press, New York, 1960), p. 252. 5 I. D. Jackson, Classical Electrodynamics (McGraw-Hill, New York, 1975), 2nd ed., p. 469. 6 N. Bloembergen, R. K. Chang, S. S. Jha, and C. H. Lee, Phys. Rev 174, 813 A96S). 7 P. S. Perehan, Phys. Rev- 130, 919 A963). 8 P. E. Eisenbergei and S. L. McCall, Phys. Rev. Lett. 26, 684 A971). BIBLIOGRAPHY Bloembergen, N., Nonlinear Optics (Benjamin, New Yoik, 1965). Bloembergen, N., Reo. Mod. Phys. 54, 685 A982).
Nonlinear Optical Susceptibilities For tower-order nonlinear optical effects, nonlinear polarizations and nonlin- nonlinear susceptibilities characterize the steady-state nonlinear optical response of a medium and govern the nonlinear wave propagation in the medium. Chapter 1 showed how the nonlinear optical response can be calculated for two model systems. Chapter 2 gives a more general discussion of nonlinear susceptibilities starting from the microscopic theory. 2.1 DENSITY MATRIX FORMALISM Nonlinear optical susceptibilities are characteristic properties of a medium and depend on the detailed electronic and molecular structure of the medium. Quantum mechanical calculation is needed to find the microscopic expressions for nonlinear susceptibilities.1 Density matrix formalism is probably most convenient for such calculation and is certainly more correct when relaxations of excitations have to be dealt with.2 Let ψ be the wave function of the material system under the influence of the electromagnetic field. Then the density matrix operator is defined as the ensemble average over the product of the ket and bra state vectors Ρ-ίΨΧΨ] B.1) and the ensemble average of a physical quantity Ρ is given by (Ρ>=(ψ|Ρ|ψ> B.2) = Tr(PP) In our calculation here, Ρ corresponds to the electric polarization. From the 13
14 Nonlinear Optical Susceptibilities definition of ρ in B.1) and from the Schrödinger equation for |ψ), we can readily obtain the equation for motion for p, known as the Liouville equation. The Hamiltonian Jfis composed of three parts, In the semicJassical approach, JPO is the Hamiltonian of the unperturbed material system with eigenstates \n) and eigenenergies E„ so that &\n) = EJn), jP^ is the interaction Hamiltonian describing the interaction of light with matter, and 3ftaaäaa is a Hamiltonian describing the random perturbation on the system by the thermal reservoir around the system. The interaction Hamiltonian in the electric dipole approximation is given by J?ml = er-E. B,5) We consider here only the electronic contribution to the susceptibilities. For the ionic contribution, we would have to replace er · Ε by - E^-R,- · Ε with g, and R, being the charge and position of the ith ion, respectively. The Hamiltoniao Ji^m, is responsible for the relaxations of material excitations, or in other words, the relaxation of the perturbed ρ back to thermal equi- equilibrium. We can then express BJ) as.3·* ■f^Jr^+^pl+i-^) B.6) with If the eigenstates \n) are now used as base vectors in the calculation, and |ψ) is written as a linear combination of |n), that is, |ψ) = Σπαπ\η), then the physical meaning of the matrix elements of ρ is clear. The diagonal matrix element p„„ = (n|pln) = |aj represents the population of the system in state |/i), while the off-diagonal matrix element pnn- = (n\p\n') = a „a*, indicates that the state of the system has a coherent admixture of |n) and |n'). In the latter case, if the relative phase of a„ and a„- is random (or incoherent), then pnn. = 0 through the ensemble average. Thus at thermal equilibrium p<°> is given by the thermal population distribution, for example, the Boltzmann distribution in the case of atoms or molecules, and pJJ = 0 for η * η'.
Density Matrix Formalism 15 We can use a simple physical argument to find a more explicit expression for (dp/dt)Kiai. The population relaxation is a result of transitions between states induced by interaction with the thermal reservoir. Let W„_„- be the thermally induced transition rate from \n) to Jn'). Then the relaxation rale of an excess population in \n) should be At thermal equilibrium we have J!p. = Y\W, p<?>, - w „.pi0|l = 0 B.9) Therefore, B.8) can also be written as B,10} The relaxation of the off-diagonal elements is more complicated.2 In simple cases, however, we expect the phase coherence to decay exponentially to zero. Then, we have, for η Φ π', = -Γηπ-Ρηη- B.Π) with T~J = Γ^1 = (Τ2)ππ- being a characteristic relaxation time between the states |n) and |«'>. In magnetic resonance, the population relaxation is known as the longitudinal relaxation, and the relaxation of the off-diagonal matrix elements is known as the transverse relaxation. In some cases, the longitudinal relaxation of a state can be approximated by lift. - Λω - -(ΓΧ'ί«,. " P^). _ B.12) Then 7Ί is called the longitudinal relaxation time. Correspondingly, T2 is called the transverse relaxation time. Thus, at least in principle, if Jf0, Jf^, and Cp/3i)relM are known, the Liouville equations in B.6) together with B.2) fully describe the response of the medium to the incoming field. It is, however, not possible in general to combine B.6) and B.2) into a single equation of motion for {P). Only in special cases can this be done. In this chapter we consider only the case of steady-state response with (P) expandable into power series of E. The tran- transient response is discussed in Chapter 21.
16 Nonlinear Optical Susceptibilities To find nonlinear polarizations and nonlinear susceptibilities of various orders, we use perturbation expansion in the calculation. Lei and <P) = (PA)) + <PB>) + ■·■ B.13) with (P<"<) = Tr(p'n>P) B.14} where p|0) is the density matrix operator for ihe system at thermal equilibrium, and we assume no permanent polarization in the medium so that (P101) = 0. Inserting the series expansion of ρ into B.6) and collecting terms of the same order with Jf^, treated as a first-order perturbation, we should obtain B.15) and so on. We are interested here in the response to a field that can be decomposed into Fourier components, Ε = S^expf/k, · r — (to, t). Then, since -^iDi = Σ.·*ΐοΐ(ωι) an^ ^mii«;)α <?,βχρ(-/ω,0' *e operator pM can also be expanded into a Fourier series pi») = Σ>">(ω) With dp<n>(Uj)/3t = -iujp^^&j), B.15) can now be solved explicitly for ρ("'(ω,) in successive orders. The first- and second-order solutions are B.16) We use here the notation Ann· = (n|^|n'). Higher-order solutions can be
Microscopic Expressions for Nonlinear Susceptibilities 17 obtained readily, although the derivation is long and tedious. Whenever diagonal elements ρ£1@) appear in the derivation, further approximation on (dPffim/tOreiin >n B.8) is often necessary in order to find a closed-form solution. We also note that the expression for ρ$(ω; + ωΛ) in B.16) is valid even for η = η' as long as «,- + «t * 0 since the term (dp$/di)relas can then be neglected in the calculation. 2.2 MICROSCOPIC EXPRESSIONS FOR NONLINEAR SUSCEPTIBILITIES The full microscopic expressions for the nonlinear polarizations (P""> and the nonlinear susceptibilities (x(ll)) follow immediately from the expressions of p(">. With Jffc, = er · Ε and Ρ = - Nei in B.14) and B.16), the first- and second-order susceptibilities due to electronic contribution are readily ob- obtained. They are given here in explicit Cartesian tensor notation: (ω - vng (ω — ω ι to + ω V η f tj + ω (ω - ω , + 'Τ-,Χ« f +/Γ-ΐ)(« η- + T„.) ,)fl-n('>)gD. ι «B'f + ίΓ,,-j,) i + w--g + 'TB.f) ! + o>n.g + iT„,s) 1 ω2 + ujt + ϊΤΛ ι 1 ω, - w„s + /T„ 1 (ω - ωππ. + ίΤππ.) \ ω2 - ω + /Γ ω, + ωπ.;
IS Nonlinear Optical Susceptibilities There are two terms in χγ> and eight terms in χ$. The calculation can be extended to third order to find x{-fk, (ω = ωι + «2 + ω,), which will have 48 terms. The complete expression for xf^, is given in the literature5 and is not reproduced here. The resonant structure of xf^kh however, is discussed in Chapter 14. In nonresonant cases, the damping constants in the denominators in B.17) can be neglected. The second-order susceptibility can then be reduced to a form with six terms, noting that the last two terms in the expression for xf/k in B.17) become (ωι ~ ω-ϊΗ«! + ««'») (ωι + "VsK^ ~ «Bf) ' With Λ' denoting the number of atoms or molecules per unit volume, the expressions in B.17) are actually more appropriate for gases or molecular liquids or solids, and p<0) is given by the Boltzmann distribution. For solids whose electronic properties are described by band structure, the eigenstates are the Bloch states, and p™ corresponds to the Fermi distribution. The expression for χί}' and %f^k should then be properly modified. Since the band states form essentially a continuum, the damping constants in the resonant denominators can be ignored. In the electric dipole approximation with the photon wavevec- tor dependence neglected, χ^\ for such solids has Ihe form3 fi2' \ }/r(q) ) where q denotes the electron wavevector, v, c, and c' are the band indices, and /„(q) is the Fermi distribution factor for the state |t\q). For condensed matter, there should be a local field arising from the induced dipole-dipole interaction. A local field correction factor L(n| should then
Diagrammatic Technique 19 appear as a multiplication factor in χ<π). We discuss the local field correction in more detail in Section 2.4. For Bloch (band-state) electrons in solids with wavefunclions extended over many unit cells, the local field tends to get averaged out, and L("' may approach 1. 2.3 DIAGRAMMATIC TECHNIQUE Perturbation calculations can be facilitated with the help of diagrams. Feynman diagrams have been used in perturbation calculations on wavefunc- tions. Here, since the density matrices involve products of two wavefunctions, perturbation calculations require a kind of double-Feynman diagram. We introduce in this section a technique devised by Yee and Gustafson.6 Only the steady-state response is considered here. The important aspects of any diagrammatic technique are that the diagrams provide a simple picture to the corresponding physical process as well as allowing one to write down immediately the corresponding mathematical expression. It is essential to find the complete set of diagrams for a perturba- perturbation process of a given order. The scheme we adopt for calculating p'"' involves in each diagram a pair of Feynman diagrams with two lines of propagation, one for the |ψ) side of ρ and the other for the (ψ| side. Figure 2.1 shows one of Fig. 2.1 A representative double-Feynman diagram describing one of the many terms in ρ'"'(ω — «! + ii>2 + ■ - ■ + un).
20 Nonlinear Optical Susceptibilities the many diagrams describing the various terms in p'n)(« = wj + ω2 + ■ ■ * + ωη). The system starts initially from |g)(g| with a population p<°>. The ket state propagates from |g) to |«') through interaction with the radiation field at wP«2 «„. and the bra state propagates from (g\ to (n\ through interaction with the field at ω3,...,ω,,_1. Then, the final interaction with the output field at ω puts the system in |«){n|. Through permutation of the interaction vertices and rearrangement of the positions of the vertices on the lines of propagation, the other diagrams for p(n) can also be drawn. The microscopic expression for a given diagram can now be obtained using the following general rules describing the various multiplication factors: 1 The system starts with \g)p{^(g\. 2 The propagation of the ket state appears as multiplication factors on the left, and that of the bra state on the right. 3 A vertex bringing \a) to \b) through absorption at ω; on the left (ket) side of the diagram is described by the matrix element with .#;„,(«,■) α e~'"'' denoted by J^ in Fig. 2.1 l*> instead of absorption, the vertex should be described by I") (\/ih)(b\ .#£, {u,)\a). Because of the adjoint nature between the bra and ket sides, an absorption process on the ket side appears as an emission process on the bra side, and vice versa.· Therefore, on the right (bra) side <b\' of the diagram, the vertices for emission If it is emission are described by - and absorption and -(l//fi)<e| •*mt (ω,-)Ι^)ι respectively. 4 Propagation from thejth vertex to the (j + l)th vertex along the \l)(k\ double lines is described by the propagator Π, = ±[ί(Σ/_!ω, - ulk + iT(jt)]"' The frequency «, is taken as positive if absorption of ω, at the ith vertex occurs on the left or emission of ω; on the right; it is taken as negative if absorption of ω; occurs on the right or emission on the left. 5 The final state of the system is described by the product of the final ket and bra states, for example, \n')(n\ after the nth vertex in Fig. 2.1 for p(n). 6 The product of all factors describes the propagation from |g)(g| to \n')(n\ through a particular set of states in the diagram. Summation of these •If the field is also quantized, Jf:ai( ω,) operating on a ket state will annihilate a photon at ω,, while if operating on a bra state it will create a photon.
\g> <ί \g> f/J (g) Fig. 2.2 The complete set of eight diagrams for the eight terms in ρί2){ω = ω,
3 2 1 l£> ' Fig· ϊ-3 The eight basic diagrams for p'"(« = W[ + ω2 + ω3).
Local FteW Coirectkm To χ"» " products over all possible sets of states yields the final result with contribu- contributions from all states. By using these rules, the diagram in Fig. 2.1 leads to the expression Ä"| _ J — 1 / \ ._l B.19) (ω, + öj + ω3 - ωήΓ + iTbc)(u1 + ω2 - ubg + iTig)(tj, - vag + iTa which is just one term in the full expression for ρ(π) (ω = ω1 + ω2 + - - ■ + ωπ). As a more concrete example, Fig. 2.2 gives the complete set of diagrams for pB) (ω = ω! + tuj) that leads to χ$ (ω = ω, + u2) in B.17). The eight diagrams (a)-(h) correspond in successive order to the eight terms in B.17). Note that xgi (ω - wL + ω2) is derived from Ji(p(T'Pi)/EJ(wl)Ek(u2). There are in fact only four basic diagrams, (a), (c), (e), and (g), in Fig. 2.2. The others can be obtained by permutation of the i^ and ω2 vortices. As another example. Fig. 2.3 presents eight basic diagrams for pC) (ω = ω1 + ω2 + ω3) that lead to χ$; (ω = ut + ω2 + ω3). There should be 48 dia- diagrams in the complete set corresponding to the 48 terms in χ(^,. The other 40 diagrams are obtained from permutations of the three vertices A,2,3) in the eight basic diagrams in Fig. 2.3. The full expression of χ|^, can then be written down from the diagrams according to the rules. What happens if identical photons appear at a number of vertices? Dia- Diagrams obtained from permutations of these vertices in a given diagram yield identical terms in p<r". They should not be discarded, and should be taken into account by a degeneracy factor attached to the terms in p(n). For example, χ8?,Cω = ω + ω + «) has 48 diagrams, but 40 of them yield terms identical to others. Thus χί?/,Cω = ω + ω + ω) has only eight terms, each having a degeneracy factor of 6. It reduces further to four terms when the damping constants in the denominators of the expression can be neglected. 2.4 LOCAL FIELD CORRECTION TO χ<π) The expressions for χ(<ι> in the previous sections are strictly correct only for dilute media. They can be written as χ<π) = JVa1"' with Ν being the number of atoms or molecules per unit volume and aln) the nth-order nonlinear polariza-
24 Nonlinear Optical Susceptibilities bilities. In condensed matter, however, the induced dipole-dipole interaction becomes important and leads to the so-called local field correction. The susceptibilities χ(η) are no longer simply proportional to α(π). The usual derivation of local field correction applies to isotropic or cubic media with well-localized bound electrons. The general theory applicable to media with any symmetry or with more freely moving electrons is not yet available. The local field at a local spatial point is the sum of the applied field Ε and the field due to neighboring dipoles ΕΛρ, Eioc-E + E^. B.20) In the Lorentz model, Edjp is proportional to the polarization; for isotropic or cubic media, it is given by7 The polarization can be expressed in terms of either microscopic polarizabili- lies and local fields or macroscopic susceptibilities and applied fields: With B.20) and B.21), the first expression in B.22) becomes B.23) } If the contribution of P"° to E^ with η > 1 is neglected [which is usually an excellent approximation since |Pl")|ll>1 ■« IP11*!], then the local field can be written as Then, from B.22) and B.23), we find B.25)
Permutation Symmetry of Nonlinear Susceptibilities and more generally χ'π)(ω = «j + ω2 + ■ ■ ■ + ωπ) ΐύτ+ --■ + w„)' Since the linear dielectric constant eA) is related to χ'1' by we can write and B.26) becomes3 B.27) with B,8) being the local field correction factor for the nth-order nonlinear susceptibili- susceptibilities. In media with other symmetry, the expression B.27) is still valid, but LM will be a complicated tensorial function of eA>(«), e'11^),..., and εο)(ωη).Β IS PERMUTATION SYMMETRY OF NONLINEAR SUSCEPTIBILITIES There is inherent symmetry in the microscopic expressions of susceptibilities. As can be readily seen from B.17), the linear susceptibility χ·]1 has the symmetry Χ!}>(ω) = χί}»·(-ω), B.29) which is actually a special case of the Onsager relation. Similarly, the nonlinear susceptibility χ$.(ω = ω1 + ωΙ) in B.17) or a similar expression for χ[^Bω = ω + ω) has the following permutation symmetry when the damping con-
26 Nonlinear Optical Susceptibilities stants in the frequency denominators can be neglected (i.e., the nonresonant cases):1·9 X$*(« = "i + «2) = Χ%(ωι = -»ι + «) -χ^(«ι-«-«,), B.30) χί^(Ζω = ω + ω) = }χ$(« - 2« - ω) - *χ$(ω = -ω + 2«). In Ihe permutation operation, the Cartesian indices are permutated together with the frequencies with their signs properly chosen. More generally, one can show that the nth-order nonlinear susceptibility also has the permutation symmetry9 Xiilj...;.*(" = ωι + «1 + ■■■ «») = xi,"j-/./(«i = ~ui ■■· + ωη + ω) B.31) If the dispersion of x<fl) can also be neglected, then the permutation symmetry in B.31) becomes independent of the frequencies. Consequently, a symmetry relation now easts between different elements of the same χ<π) tensor, that is, xi.1?,,. .1 remains unchanged when the Cartesian indices are permuted. This is known" as Kleinman's conjecture,10 with which the number of independent elements of χ(η> can be greatly reduced. For example, it reduces 27 elements of Xt2> to only 10 independent elements. We should, however, note that since alt media are dispersive, Kleinman's conjecture is good approximation only when all frequencies involved are far from resonances such that dispersion of χ(η> is relatively unimportant. 2.6 STRUCTURAL SYMMETRY OF NONLINEAR SUSCEPTIBILITIES As optical properties of a medium, the nonlinear susceptibility tensors should have certain iorms of symmetry that reflect the structural symmetry of the medium. Accordingly, some tensor elements are zero and others are related to each other, greatly reducing the total number of independent elements. As an illustration, we consider here the second-order nonlinear susceptibility tensor Xln- Each medium has a certain point symmetry with a group of symmetry operations {S}, under which the medium is invariant, and therefore χ1 remains unchanged. In real manipulation, S is a second-rank three-dimen- three-dimensional tensor Slm. Then, invariance of χ<2) under a symmetry operation is
Table 2.1 Independent Noimmishing Elements of χ(ίι(ω = ω, + ω2) for Crystals of Certain Symmetry Classes Symmetry Class Independent Nonvanishing Elements Triciinic 1 Monoclinic 2 Qrthorhombic 222 mm! Tetragons! 4 422 4mm 32m Cubic 432 43m 23 Trigonal 3 32 3m Hexagonal 6 622 6 mm All elements are independent and nonzero xyz, xzy, xxy, xyx, yxx, yyy, yzz,yzx, yxz, zyz, izy, ixy, zyx (two fold axis parallel to^) xxx, xyy, xzz, xzx, xxz,yyz, yzy, yxy, yyx, zxx, zyy, zzz, zzx, zxz (mirror plane perpendicular toy) xyz, xzy, yzx, yxz, zxy, zyx xzx, xxz, yyz, yzy, zxx, zyy, zzz xyz ** -yxz, xzy = -yzx, xzx = yzy, xxz — yyz, zxx = zyy, zzz, zxy = -zyx xyz - yxz, xzy — yzx, xzx — -yzy, xxz = -yyz, zxx — -zyy, zxy = zyx xyz - -yxz, xzy = -yzx, zxy = -zyx xzx — yzy, xxz = yyz, zxx = zyy, zzz xyz — yxz, xzy = yzx, zxy = zyx xyz - - xzy — yzx = -yxz — zxy = - zyx xyz — xzy = yzx = yxz — zxy — zyx xxx - -xyy - -yyz = -yxy, xyz = -yxz, xzy = -yzx, xzx =yzy,xxz =yyz,yyy = —yxx = -xxy = -xyx, zxx = zyy, zzz, zxy — —zyx xxx — -xyy - —yyx ™ -yxy, xyz = -yxz, xzy = -yzx, zxy — —zyx xzx =- yzy, xxz — yyz, zxx — zyy, zzz, yyy = -yxx = - xxy — - xyx (minor plane perpendicular to it) xyz — — yxz, xzy — — yzx, xzx = yxy, xxz = yyz, zxx = zyy, zzz, zxy = -zyx xxx - -xyy - -yxy - -yyx,yyy ~ -yxx — -xyx = —xxy xyz = — yxz, xzy = -yxz, zxy — -zyx xzx = yzy, xxz = yyz, zxx = zyy, zzz yyy — -yxx — -xxy = —xyx
IS Nonlinear Optical Susceptibilities explicitly described by B.32) For a medium with a symmetry group that consists of η symmetry operations, η such equations should exist. They yield many relations between various elements of χ'2', although often only a few are independent. These relations can then be used to reduce the 27 elements of χ(ϊ) to a small number of independent ones. An immediate consequence of B.32) is that χB) = 0 in the electric dipole approximation for a medium with inversion symmetry: with S being the inversion operation, S · i = -e, B.32) yields χ^ί = —xiyi = 0- This explains why χ'2> for a free electron gas does not have an electric dipole contribution as shown in Chapter 1. Among crystals without inversion symmetry, those with the zincblende structure such as the III-V semiconductors have the simplest form of xm. They belong to the class of TdD3m) cubic point symmetry. Table 2.2 Independent Nonvanishing Elements of x^'iw = ut + ω2 + ι^) for Crystals of Certain Symmetry Classes Symmetry Class Independent Nonvanishing Elements Tridinic All 81 elements are independent and nonzero Tetragonal xxxx = yyyy, zzzz, 422,4mm, yyzz = zzyy, zzxx - xxzz, xxyy = yyxx,yzyz " zyzy, 4/mmm,42m zxzx = xzxz,xyxy = yxyx, yzzy — zyyz,zxxz = xzzx, xyyx = yxxy Cubic xxxx = yyyy = zzzz,yyzz - zzxx =- xxyy, 23, mi zzyy = yyxx = xxzz, zyzy - xzxz - yxyx, yzyz = zxzx — xyxy, zyyz = xzzx - yxxy, yzzy = zxxz = xyyx 432,43m, m3m xxxx — yyyy — zzzz yyzz = zzyy - zzxx - xxzz - xxyy = yyxx yzyz = zyzy — zxzx = xzxz = yxyx = xyxy yzzy = zyyz = zxxz - xzzx = xyyx = yxxy Hexagonal zzzz, xxxx = yyyy — xxyy + xyyx + xyxy 622,6mm, xxyy ■= yyxx, xyyx — yxxy, xyxy =* yxyx, d/mmm, 6m2 yyzz - atjw;, izjy - zzxx, zyyz = ϊχχϊ, yzzv — Ar/ix, >>zyi = λγγλζ, zyzy " zxzx Isotropie xxxx — yyyy = zzzz, yyzz — zzyy — zzxx — xxzz = xxyy = yyxx, yzyz — zyzy = ζκχ — λϊλ^ — χκχκ *" yxyx, yzzy — zyyz — zxxz — xzzx = xyyz = yxxy, xxxx — xxyy + xyxy + xyyx
Practical Calculations of Nonlinear Susceptibilities 29 Although many symmetry operations are associated with 7^ D3m), only the 180° rotations about the three four-fold axes and the mirror reflections about the diagonal planes are needed to reduce χ<2). The 180° rotations make xffi = -xffl = 0, xjg = -x|g = 0, and χ$ = -χ$ = 0, where ί,;, and k refer to the three principal axes of the crystal. The mirror reflections lead to the invariance of xfyii Φ j Φ k) under permutation of the Cartesian indices. Consequently, xjjj,(' * j * k) is the only independent elemen! in χB) for the zincblende crystals. For other classes of crystals, the forms of χB) can be similarly derived through the corresponding symmetry operations. The symmetry consideration here is the same as the one used to derive the electrooptical tensor [which is actually a special case of χα'(« = w, + ω2) "witli ω2 = 0] and the piezoelectric tensor.11 The forms of χ<2) for second-harmonic generation are in fact identical to the latter.*3 We reproduce a part of χB>(ω = ω, + ω2) for various classes of crystals in Table 2.1. The above symmetry consideration for χB) can of course be extended to higher-order nonlinear susceptibilities. In particular, symmetry forms for χC) are most important in view of the many interesting third-order nonlinear optical effects that can be observed readily in almost all media. Table 2.2 lists the χC) tensors for the more commonly encountered classes of media.12 2.7 PRACTICAL CALCULATIONS OF NONLINEAR SUSCEPTIBILITIES Symmetry operations drastically reduce the number of independent elements in a nonlinear susceptibility tensor, but then for a given medium, we would also like to know the values of these independent elements. While they can often be measured (see, for example, Section 7.5), it is also important that they can be calculated from theory. A successful theoretical calculation can help in predicting χ'"' for media not easily subject to measurements or for the design of new nonlinear crystals. In principle, the microscopic expressions, such as the one for χ^ in B.17), with appropriate local-field correction, can be used for such calculations. However, in most practical cases these expressions are useless because neither the transition frequencies nor the wavefunctions for the material are sufficiently well known. This is especially true for large molecules or solids. Simplifying models or approximations often are needed. If all frequencies involved are far from resonances, one simplifying assumption often used is to replace each frequency denominator in the microscopic expression of χ<"> by an average one and bring all frequency denominators out of the summation [see, for example, χB) in B.17)]. Then the summation over matrix elements can be greatly simplified through the closure property of the eigen- states and can be expressed in terms of moments of the ground-state charge distribution. The problem reduces to finding the ground-state wavefunction of the system.13
30 Nonlinear Optical Susceptibilities The foregoing approximation, however, is too drastic to yield good results. A more successful calculation of χ'π> can be done by the bond model. Such a model was used in the early 1930s to calculate the linear polarizability of a molecule or the linear dielectric constant of a crystal.14 The bond additivity rule was assumed: the induced polarizations on a molecule (or a crystal) is the vector sum of the polarizations induced on all bonds between atoms. In other words, the bond-bond interaction is neglected. The same rule can be used in the calculations of χ1. We can write ΧΜ-Σ*Ρ B-33) κ where aj?' is the nth-order nonlinear polarizability of the K\h bond in the crystal (or medium), and the summation is over all the bonds in a unit volume. Thus, with known crystal structure, the calculation of χ* reduces to the calculation of a^1' for different types of bonds. We discuss here only the calculations of χ<2), using the zincblende crystals as an example. The general procedure is as follows. The linear bond polariza- polarizability a^' is first calculated as a function of the applied field using the recently well developed bond theory.11 The second-order nonlinear bond polarizability αψ is then obtained from the first derivative of ajj' with respect to the applied field. Finally, the summation of B.33) over the bonds is performed to find χ'2'. We assume here that a simple crystal can be constructed entirely out of the same type of bonds, and the bonds are cylindrically symmetric. The linear susceptibility χ<|> of the crystal can then be written as + G<V]> B.34) where ajj1' and αψ are the polarizabilities parallel and perpendicular lo the bond, μ = u^/a^K and Gjj1* and G^' are the respective geometric factors arising from the vectorial summation over the bonds. Both Gjj" and G^' are proportional to the number of unit cells per unit volume. For the zincblende structure, G'11 = $G<P = 4JV/3, and B.34) becomes χα>-<£A + 2μ)«ί». B.35) The next step is to find an approximate expression for «j1' through χ£'. The microscopic expression of χ'}' in B.17) away from resonance has the form
Practical Calculations of Nonlinear Susceptibilities 31 In the low-temperature limit, pf = 0 for all states except the ground state. Then, following the approximation of replacing ωΠί in the denominator by an average ω and the sum rule16 B.36) reduces to with Ωρ = 4-nNe2/m being the electron plasma frequency. This simplified expression for xj)> has actually been shown more rigorously by Penn for solids in the limit of zero frequency.17 From B.34), we now have B.39) We are, however, interested in af1 as a function of applied field. The polarizability should depend on the field through the field perturbation on the transition frequencies and matrix elements. However, in the approximate form of B.39), ajj" can depend on the field only through wjjr To find an repression for ülg is where the bond theory comes in. Physically, hu„g = E? can be regarded as an average energy gap between the filled and unfilled states. It can be written as15 E,= [Ei + C>]1/2 B.40) where EA and C are known as the homopolar and heteropolar gaps, respec- respectively, and, in the bond theory, have the expressions Ci1 * ad" and B.41) In these expressions, a, b, and s are constant coefficients, ZA and ZB are the valences, and rA and rB are the covalenl radii of the A and Β atoms forming the bond, d — rA + rBis the bond length, and exp(—k^d/2) is the Thomas-Fermi screening factor. If A and Β are identical atoms, then C = 0. Equation B.40) can be derived easily from molecular orbital theory.18 The bond electrons have two eigenstates, a bonding state and an antibonding state. The energy dif-
32 Nonlinear Optical Susceptibilities ference between the two states is Eg. For a homopolar bond {A = ß), the bond electrons see a symmetric potential with respect to the bond center and Eg = E„. For a heteropolar bond {Α Φ Β), the bond electrons see an anti- syymmetric potential, and Eg = El + C1 with C proportional to the asymmet- asymmetric part of the potential. The wavefunctions of the bonding and antibonding States along the bond are shown in Fig- 2.4. It is seen that in the heteropolar case, there is a charge transfer from the side of the less electronegative atom to the side of the more electronegative atom. According to the molecular orbita) theory, the amount of transferred charge Q is related to the heteropolar gap C by "-'f B.42) Figure 2.4 also shows that there is a bond charge cloud between the two atoms. The magnitude of the bond charge derived from the bond theory is 2eE 11 + B.43) Fig. 2.4 Sketches of electronic wavefunctions of (u) the bonding stale and (b) the antibonding state along the bond connecting the atoms A and B. The solid curves are for the homopolar case and the dashed curves for the heteropolar case.
Practical Calculations of Nonlinear Susceptibilities 33 Levine1' suggests that the bond charge may be considered as a point charge sitting at distances rA and rB, respectively, from atoms A and B. We can now discuss how the bond polarizabih'ty changes when the bond is subject to an external field. The change occurs through the field perturbation on the charge distribution. In our description here, ajj1' depends on the applied field Ε through the dependence of Ε on Ε while EA and C depend on Ε through field-induced changes in the charge transfer and bond charge. However, since the applied field is not expected to change the bond length, we have SE^dEj = 0 from B.41). The second-order nonlinear bond polarizability afj£ is obtained from da^/ΘΕ^.. If I and ή denote the two directions parallel and perpendicular to the bond, respectively, then from ihe symmetry argument, only affa and ot^1,, are nonvanishing. We also neglect o^ by assuming that a field transverse to the bond will not significantly perturb the charge distribution. Thus af^ is the only nonvanisliing element of a™. Using B.39) and B.44), we find dB( -2h2ttjC gc Now, either B.41) or B.42) can be used to calculate dC/d£k. The two, however, correspond to two different physical pictures. In B.41), the applied field changes rA and rB, but keeps rA + rB = d. In terms of the simple mode! where the bond charge can be treated as a point charge sitting at distances rA and rB away from the atoms A and B, the field then simply shifts the position of the bond charge along the bond. This is known as the bond-charge model." In B.42), on the other hand, it is the field perturbation on the charge transfer Q that relates C lo the field. This is the charge-transfer model.20 The bond-charge model involves, with Ar = ArA = - irfi, and since qbr= ajjl)(io')A£{(w') for ω' -* 0, we find from B.41), B.45),
34 Nonlinear Optical Susceptibilities B.46), B.34), and B.38) B.47) The charge-transfer model following B.42) gives It is assumed in this model that the field-induced charge transfer is from atom Β to atom A, treating the atoms as points. Since a$\af)&Et(<ar) = t^Qd, we have from B.45) and B.48) B-49) We should, however, keep in mind that the description of how an applied field modifies the charge distribution in both models is still fairly crude. In reality, the electronic charges are broadly distributed in the region between the two atoms. The peak of the distribution is near the center of the bond. As an example, a contour map of the valence electron distribution around a Ga-As bond obtained by empirical pseudopotential calculation is shown in Fig. 2.5.21 In the presence of a dc external field along the bond, the charge distribution becomes only slightly more asymmetric with its peak essentially unshifted. This is shown in Fig. 2.6 for the charge distributions along the Si-Si and Ga-As bonds.22 The field-induced shift of the bond charge in the bond-charge model actually refers to the shift of the center of gravity of the valence electron distribution, while the field-induced charge transfer in the charge-transfer model refers to the redistribution of the valence charges around the bond from one side of the bond center to the other. Finally, we can obtain χ{^ of a given medium from «^ for various bonds, where i,), and k denote the three orthogonal symmetry axes in the crystal:
Practical Calculations of Nonlinear Susceptibilities Fig. 2.5 Coniour map of valence electron density distribution (in units of e per primitive cell) forGaAsin the A, -1,0) plane. (From Ref. 21.) where [G^)iJk is a geometric factor for the Λ-type bonds reflecting the structure of the medium. We note that with («${)a expressed in terms of χ!,-1 rather than af1 in B.47) and B.49), even the total field correction has been somehow taken into account in the above derivation. We now use InSb as an example to illustrate the calculation of χ$. The crystal has a zincblende structure; therefore, the only nonvanishing elements of χB) are x<?j{ with i*j* k. There is only one type of bond in the crystal: those connecting In and Sb. The geometric factor Gj$t is then given by 4^/3/3 and the density of unit cells JV is related to the bond length d by Ν = 3vT/16i/3. We also have Gj'> = \ΰψ = 4JV/3. From B.47), B.49), and B.50), the bond- ■(lill -MID Fig. 2.6 Sketches of the charge distribution along a bond in (a) Si and (t) GaAs. Solid and dashed curves refer to cases with and without an external field along the bond, respectively. (Courtesy of S. Louie.)
36 Nonlinear Optical Susceptibilities charge model gives , , _ 32«f'c[X<»(«)]W)] [g, g and the charge-transfer model gives (vß) \ = "Ί. li* 'ω^ '* '" '' ij S2\ We calculate here χψγι in the low-frequency limit ω ~ ω' - 0. For InSb, d = 2.5 Ä, Es = 3.7 eV, £k = 3.1 eV, C = 2.1 eV, χA> - 1.17 esu, ΛΩ - 13 eV, ZA = 3, ZB = 5, /> = rfl = ti/2, ftexp(-M/2) = 0.12 e2, μ = {, and « = 0.6 e,23 we obtain (x^)BC= 1.6 x Wesuandta^cT.« 2.3 Χ ΙΟ esu. The results of both models are in fair agreement with the experimental value of χ^; = C.3 ± 0.7) χ 10"* esu. This should be considered satisfactory in view of the crude approximations in the models. The calculations can also be extended to higher-order nonlinear susceptibili- susceptibilities. However, because of the crude approximations involved, they become much less reliable. Also, since we use the covalent bonding picture in the models, the calculations are less suitable for ionic crystals. In nonlinear optics, we are often interested in materials with high nonlinearity. This discussion suggests that the materials should have high nonlinearity in bond polarizabili- ties. For large χB), the crystal structure should also be as asymmetric as possible so that there is a minimum of vectorial cancellation in summing over αψ of all bonds. The calculations here are good only in the low-frequency limit. The ap- approximations in the models break down when the optical frequencies are close to the absorption bands. Because of resonant enhancement, the transitions with transition frequencies closer to the optical frequencies contribute much more to the susceptibilities. In order to calculate χ'"* and its dispersion in these cases, we must use the full microscopic expression of χ{ΐ" such as those derived in Section 2.2. Then detailed information about the transition matrix elements and frequencies of the material is necessary. Such calculations have been carried out by several authors on χ<2)Bω) of zincblende semiconductors with various degrees of approximation. In most cases, constant matrix ele- elements are assumed. The more accurate calculations, however, are those with wavefunctions and energies of the band states derived from the empirical pseudopotentia! method,24 which has been extremely successful in reproducing χA|(ω) for zincblende semiconductors; it should therefore also yield accurate results for χB)Bω). An example is shown in Fig. 2.7 for InSb. The peaks and shoulders in the spectrum generally correspond to resonances of ω or 2 ω with the critical point transitions. The results also show that it is important to
Miller's Coefficient Λ-Κ3-5) ■Γ{4-5) A-LD-5) A-LC-5) 800 600 Fig. 2.7 Dispersion of χ^;Bω) of InSb calculated using the empirical pseudopoten- tial method. The pealcs arise from inteiband transitions in the regions indicated. (From Ref. 24.) include the dispersive effects of both the matrix elements and the density of states for transitions in the calculations. Full quantum mechanical calculations of χB) of B.17) for molecular crystals have also been carried out using semiempirical Hartree-Fock LCAO (linear combination of atomic orbitals) methods by many researchers.25 They were able to predict quite satisfactorily the measured values of χ<2). Highly asym- asymmetric molecules with strong charge-transfer bands appear to yield large |χ<3)| if the crystal structure is also highly asymmetric. 2.8 MILLER'S COEFFICIENT Miller defined a coefficient26 B,53) and found empirically that iijk has only weak dispersion and is almost a constant for a wide range of crystals. This is known as Miller's rule. It suggests that high refractory materials should have large nonlinear susceptibilities. The weak dispersion of i,Jlt can be seen from either the bond-charge or the
38 Nonlinear Optical Susceptibilities charge-transfer model. Equations B.51) and B.52) show that for ω' -» 0, Δ,-yjt = constant independent of frequencies. The constant is, however, proportional to the heteropolar gap C, and does change, although only mildly, from crystal to crystal. That the measured Δ,^. is indeed proportional to C for a large number of semiconductors has been demonstrated by Levine." For a crystal with several different types of bonds, a weighted average C must be used. The values of i,Jlt for most nonlinear crystals are around few times 10 ~6 esu. 2.9 CONVENTIONS ON NONLINEAR SUSCEPTIBILITIES The definitions of nonlinear susceptibilities in the literature vary and have caused some confusion. This section clarifies the conventions used in this book. The definition of nonlinear susceptibilities is governed by the following rela- relation between the nonlinear polarization P1"' and the electric fields E;: Ρ<">(ω) = χ(η)(ω = ux + ω2 + ■ ■ ■ + «„) :Ε1(ω1)Έ,2(ω2) ■ ■ ■ Επ(ωπ) B.54) with Ε,- and Ρ(π) expressed as complex quantities: E, = <ffexp(ik;-r- iwji P«0(w) . ^-O assuming ω; and ω are both nonzero. Many authors have written the ampli- amplitudes of E; and P(n) in somewhat different forms with Ey = ii/exp(fk, · r - ίω,-ί) B.56) Ρ(π>(ω) - iä"<">exp(ik ■ r - lut) and defined a nonlinear coefficient d<n) to connect the amplitudes 5"<") = d(n>:^i-·- *.' B.57) or pt-) = B)"d("): E!E2 - ■ ■ En. B.58) Comparison of B.54) and B.58) gives
Conventions on Nonlinear Susceptibilities 39 and in particular d#l = hxfjk- Equation B.59), however, needs modification when there are dc fields present. For ω, = 0, the corresponding dc field E, should be related to < and £,' by E, = l€i = £[. Then, if s of the η fields, namely, Ε^.,.,Ε^ are dc, we have, following B.54) and B.57) as definitions for χ'"' and d(">, p(») = x(»>:E1 ■■■ Ε,(^)""ν;+1 ■■■ ^exp[i(kJ+1 + ■■■ +kj-r r , . ^ " B-60) - id': E, ■ ■ ■ E, //+1 ■ ■ ■ <Jap[/(kI+l + ■ ■ ■ + kj · r and hence a<»)= B)-"+ι+!χ('·>, B.61) More explicitly, B.54) takes the form = Σ x'i"X..j,(u = ui + U2+ ■■■+«j /„ /,...,;. B.62) Our Convention is that the term χί/"ί,..^(ω = wj + ι»2 + ■■■ + ωπ) £ί](ω1)£1/ΐ(ω2) ■ ■ ■ £,,(«„) can be written with the fields arranged in any order as long as the subiadices of χ<π) are arranged in the same order, but no additional contribution to ?/"> should arise from permutation of the fields in B.62). The conventional notation demands that the field arrangement should always follow the ordering of the frequencies in the argument of χ'. This leads to the question of what happens if two or more fields involved have the same frequency. In our convention, permutation of the fields with the same frequency should yield no additional contribution to /*"'. For example, we have for second-harmonic generation. B.631 Φ χ%Ε()Ε() £»Ε>) In the convention using the d coefficients, however, all terms derived from permutation of the fields with the same frequency must be included in the expression of the nonlinear polarization. For example, B.64)
40 Nonlinear Optio] Susceptibilities In comparison with lif'K - ω, + ω2) = rf«, (ω, = u, + «3 )£,(«,) E>;) B,65) we notice that since the nonlinear response of the medium is not expected to have a sudden change as ωλ approaches ω2, the coefficient ά^(ω3 = «j + «3> should change smoothly to 2α®ζ(ω3 = 2ω,). The result that Κ$(ω3 = ωι + ω3)]ω,-ω, = 2ΛΡ,[BωΙ) with j # k has caused a great deal of confusion. A similar situation occurs when one or more fields have their frequencies ap- approach zero, as we discussed earlier. Our convention here avoids such diffi- difficulty: χ!;1(ω3 = ω, + ω2) changes continuously to χ^(ω3 = 2ΐύΎ) as ω! ap- approaches ω3, or to X^C«! = 0 + 10!) as ω2 approaches zero. The continuous variation of χ^ with frequencies can be explicitly seen in the microscopic expression of xjyj, in B.17). Another convention proposed by Maker and Terhune27 and often used for third-order nonlinearity is to indicate explicitly the number of terms one can obtain by permutation of different field components in the expression of the nonlinear polarization. For example, we write u3) = Σ Djk J.k.l where D/kl is the degeneracy factor for the particular terms. If £,(ω[) Φ Ek(u2) Φ £/(«3), then Djkl = 6, indicating that six terms can be obtained by permut- permuting the three fields. For £,(«]) = Ek(u2) Φ E,(u}), we have DJk, = 3, and for £,(«,) = E/t(«2) = E,(u3), we have Djkl = 1. This convention also has the difficulty that the nonlinear coefficients C^.t{a - uL + ω2 + ω,) vary discon- tinuously as the frequencies become degenerate. Further discussions of nonlinear optical susceptibilities appear in later chapters in connection with the specific nonlinear optical problems discussed. REFERENCES 1 I. A. Annstrong, R Bloembergen. J. Duelling, and P. S. Peishaa, Phys. Rev. 127,1918 A962). 1 N. Bloembergen and Y. R. Shea, Phys. Rev. 133, A37 A964). 3 N. Bloembergen, Nonlinear Optics (Benjamin, New ΥοΛ, 1%5). 4 C. P. Slichter, Principles of Magnetic Resonance, 2nd ed. (Springer-Verlag, Berlin, 1978), Chapter 5. 5 N. Bloembergen, H. Lotem, and R. T. LuDch, Indian J. Pure Λρρί. Phys. 16,151 A978). 6 T. K- Yee and T. K. Gusiafson, Phys. Rev. A18,1597 A978). 7 See, for example, C. Kind, Introduction to Solid State Physics, 5ih ed. (Wiley, Ne» York, 1976), p. 406. 8 D. Bedeaux and N. Bloembergen, Physica (Amsterdam) 69, 67 A973).
Bibliography 41 9 Υ. R Shen, Phys. Rev. 167, 818 A968). 10 D. A. KJeinman, Phys. Rev. 126,1977 A%2). 11 See, for example, J. F. Nye, Physical Properties ofCrystals (Oxford University Press. London, 1957). 12 P. N. Butcher, Nonlinear Optical Phenomena (Ohio Stale University Press, Columbus, 1965), pp. 43^50. 13 F. Ν. Η. Robinson, Bell Syst. Tech. J. 46, 913 A967); J. Phys. C 1, 286 A968); S. S. Jha and N. Bloembergen, Phys. Reu. 171, 891 {1968); C. Flytzaois and J. Duelling, Phi's. Rev. 178, 1218 A969). 14 K. G. Denbigh, Trans. Faraday Soc. 36, 936 A940). 15 See, for example, J. C. Phillips, Covalent Bonding in Crystals, Molecules, and Polymers (University of Chicago Press, Chicago, 1969); Bonds and Bands in Semiconductors (Academic Press, New York, 1973). 16 The relation is known as the Thomas-Reiche-Kubn sum rule in solid slate physics. See, for example,.!. Ziman, Principles of the Theory of Solids (Cambridge University Press, Cambridge, 1965), p. 224. 17 D. R. Penn, Phys. Rev. 128, 2093 A%2). 18 See, for example, C. A. Coulscm, Valences (Oxford University Press, London, 1961). 19 B. F. Levine, Phys. Rev. Lett. 22, 787 A969); Phys. Reu. B7, 2591 A973); 2600 A973). 20 C. L. Tang and C. Flytzanis, Phys. Ren. B4,2520 A971); C. L. Tang, IEEEJ. Quant. Electron. QE-9, 755 A973); F. Scholl and C. L. Tang, Phys. Ren. B8, 4607 A973). 21 J. P. Walter and M. I. Cohen, Phys. Rev. Lett. 26,17 A971). 22 S. Louie and M. L. Cohen, personal communication. 23 Values of the various quantities are obtained from J. C. Phillips and J. A. Van Vechien, Phys. Rev. 183. 709 A969). Μ C. Y. Fond and Y. R, Shen, Phys. Rev. B1I, 2325 A975). 25 See, for example, J. L Oudar and J. Zyss, Phys. Ret>. A26, 2106 A982); J. Zyss and I. L. Oudar, Phys. Rev. AM, 2028 A982); C. C. Teng and A. F. Garito, Phys. Rev. Lett. 50, 350 A983); and references therein. 26 R. C Millet, Appl. Phys. Leu. 5,17 A964). 27 P. D. Maker and R. W. Terhune, Phys. Rev. 137, A801 A965). BIBLIOGRAPHY Bloembergen, K, Nonlinear Optics (Benjamin, New York, 1965). Butcher, P. N., Nonlinear Optical Phenomenon (Ohio Slate University Press, Columbus, 1965). Duelling, J., and C. Fiytzanis, in F. Abetes, ed.. Optical Properties of Solids (North-Holland Publishing Co., Amsterdam, 1972), p. 859. Flytianis, C in H. Rabin and C. L. Tang, eds., Quantum Electronics (Academic Press. New York, 1975). Shen, Y. R., in N. Bloembergen, ed., Nonlinear Speclroscopy, Proceedings of Ihe International School of Physics, Enrico Fermi, Course LXIV (North-Holland Publishing Co., Amsterdam, 1977), p. 17ft
General Description of Wave Propagation in Nonlinear Media Waves can interact through nonlinear polarization in a medium. Propagation of waves in the presence of wave interaction leads to various nonlinear optical phenomena. The quantitative description of a lower-order nonlinear optical effect usually starts with a set of coupled wave equations with the nonlinear susceptibilities acting as the coupling coefficients. This coupled-wave approach can also be generalized to include waves other than electromagnetic. This chapter is devoted to a general discussion of coupled electromagnetic waves in a medium and the solution of the coupled wave equations under certain approximations. Applications of the analysis to specific nonlinear optical phenomena appear in later chapters. 3.1 COUPLED WAVES IN A NONLINEAR MEDIUM The wave equation that governs optical wave propagation in a medium is 0-1) which follows directly from the Maxwell equations A.5). Wave interaction gives rise to the nonlinear terms in P. We assume that both E(r, () and P(r, /)
Coupted Waves in a Nonlinear Medium can be decomposed into a set of infinite plane waves: E(r, t) = £E,(k„ «,) -Σχ'Ή-,ϊ'Ε,Οί/.ωί). C.2) and where ^, is taken as essentially independent of time. With είω,) = 1 4ττχ<1|(ω,), C.1) becomes [v X(V X) -^ C.3) Suppose PNL(km, ω) = P(a>(km, ω) is a nonlinear polarization from the prod- product of E;(k„ vj ■·■ E„(k„, «„). Then, for the η fields E,(kt, ω,), there should be η corresponding wave equations similar to C.3). Together with C.3), they form a set of (n + 1) coupled wave equations. Note that while um should be equal to ω in PNL(km,tJm) because of photon energy conservation in the steady-state case, km need not be exactly equal to k since wave momentum conservation is not strictly required in a finite medium. Equation C.3) clearly indicates that the various waves E,(k,,«,) arc nonlinearly coupled through the nonlinear polarization Ptn-, and their propagations in the medium will conse- consequently be very different from the linear case where PNL = 0. Through nonlinear coupling, energy can now be transferred back and forth between waves, and the larger PNL is, the more pronounced the effect should be. The coupled wave approach was first used in the description of microwave para- parametric amplification1 and later adopted by Armstrong et a!.3 for describing wave interaction in nonlinear optics. The simplest case of optical wave interaction deals with second-order nonlinear optical effects. We use it here as an example to illustrate the coupled
44 Genera] Description of Wave Propagation in Nonlinear Media wave formalism. Consider three waves Wku ut), E{k2, ω2), and E(k, ω = Wj + ω2) interacting in a medium by second-order nonlinear polarization. Then, the coupled wave equations from C.3) are ω) : E£(k2, W;)E (k, «), £»,B)/,, —,,_,. ^ ■ E'f'L *.\tr*/L· ** \ — ΙΧ' \^2 — W — ίύ-,) . ΕΊ,Κ, ίύ Jü^rtKi, tOi I <r and I V x(v As seen here, the nonünear susceptibilities appear explicitly as the coupling coefficients. They determine the rate of energy transfer among the three waves. In the case of a dissipationless medium, the permutation relation χ$.*(ω = ω, + «j) = Χ%(ωι = ~ω2 + ω) = xi?j(«2 = ω — ω1) exists (see Section 2.5). This is actually a necessary condition for the coupled wave equations to satisfy the requirement that the total energy in the three waves is a constant, as we shall see in Section 3.2. The photon energy and momentum conservations in the present case are ω = ωλ + ω2 and k = kL + k2, respectively. For most effective energy transfer among the waves, one naturally expects that both photon energy and momentum conservations should be satisfied in the wave interaction. Therefore, even though k = ^ + k2 is not required, as mentioned earlier, satisfaction of the relation is preferred for the maximization of the wave coupling. This photon momentum matching condition is known in nonlinear optics as the phase matching condition, and it is one of the most important considerations in many nonünear optical processes. Detailed solu- solutions of C.4) appear in later chapters. 3.2 FIELD ENERGY IN A NONLINEAR MEDIUM The Maxwell equations A.5) lead to the following familiar energy relation for the fields: Äv-(EXB)=-^|(£2 + ^)-E-f. C.5)
Field Energy in a Nonlinear Medium 45 With Ε Χ Β being the Poynling vector, it shows that the rate of electromag- electromagnetic energy flowing out of a unit volume is equal to the reduction rate of the stored electromagnetic energy density. If the dispersion of the medium can be neglected, then the polarization Ρ can be written as C.6) and C.5) reduces to the form with U(r, ή = —(Ε1 + Β2) + ^Ε-χ(Ι)·Ε + |ε·χ<2>: ΕΕ + - ■ ■ C.8) being the instantaneous electromagnetic energy density. This is clearly not valid in a medium with dispersion, since the susceptibilities are defined only in terms of the Fourier components of fields and polarizations. In reality, it is more meaningful to consider the time-averaged energy relation. Let us il- illustrate the problem by first assuming a linear medium. We begin by assuming a quasi-monochromatic field where 4f(l) is a slowly varying amplitude. Expressing &(t) as a Fourier integral, we have E(r, 0 = \d-qS{u + η)β<-»··--')-"i' + c.c. C.10) Then the linear polarization takes the form Pro(r, 0 = [αηχA){<ύ + η)·^(ω + η)ε'»-'-"^-'ψ + c.c. C.Π) This leads to , 0 = /</n(-/)(« + 1)|X(» + ^rn + ■ ■ - | -<?U + ij) C.12) "ω" + c.c.
4ΐ General Description of Wave Propagation in Nonlinear Media With xg-'(w) - χ$'<-«)*, the time average of C.5) yields3 (£v(ExB))--!<i/m>-ß C.13) where ß = ^('W'-<?@. C.14) and e = e' + ίε"= 1 + 4ΐ7χA). As may be inferred from the preceding equations, (Ua>) is the average field energy density stored in the linear medium and Q is the average power density dissipated into heat through absorption because e" Ψ 0. Equation C.13) is therefore an energy conservation relation one should expect physically. The above calculation can be extended to the nonlinear case.4 Additional terms in the energy relation are expected from wave coupling. Consider, for example, three waves with ω3 = ωι + ω2 and k3 = kL + k2 interacting in a nonlinear medium via χ11'. One finds, with the help of the permutation symmetry relation of xra, that the average field energy density has an additional term Xw(«i - -«! + «,): *ί(* W) obtained from /-I Note that only when \umdxU)/d<jim\ « |χB)| can we write (Um) = 2&* ■ Xra("i = ~ωι + «3> : ^2^3 + c-c- C-16) The last equation, however, is frequently used in the literature5 to describe the interaction free energy density for the wave coupling. One would write the free energy density as Fm = - (UB)) using C.16) and derive the nonlinear polari- polarization from PB)(lü„,) = — \dF{2)/d&*. From
Slowly Varying Amplitude Approximation 47 11· d<f3 = 3Ρ(ί>*(ω3)/3/ί W2*, the permutation symmetry rela- relation of χ|2> immediately results. Although this is a practice one can indeed follow, we must realize that it is actually the permutation symmetry of χ<3) that leads to the expression of (Ua>) in C.15). Conversely, it is the existence of (Um) that physically justifies the permutation symmetry of χ|2). Near reso- resonances, when dissipation in the medium becomes important, the permutation symmetry relation of χB) breaks down, and accordingly, C.15) is no longer valid. More generally, in a nonabsorbing medium, the time-averaged field energy density should be4 (U) = £ (Vw) C.17) where (t/1"') arises from nonlinear coupling of (η + 1) waves via then ih-order nonlinearity in the medium, and is given by ' 3«, ' * 2 The time-averaged energy conservation relation takes the form 33 SLOWLY VARYING AMPLITUDE APPROXIMATION In actually solving the coupled wave equations, several simplifying approxima- approximations are often made.2· 6 Among them are the slowly varying amplitude approximation, the infinite plane-wave approximation, and constant pump intensity approximation. We discuss here only the slowly varying amplitude approximation and leave the others to later chapters. As mentioned earlier, wave coupling in a nonlinear medium results in energy transfer among waves. Therefore, the wave amplitudes are expected to change in propagation. We assume for illustration a plane wave propagating along ί: Since the energy transfer among waves is usually significant only after the waves travel over a distance much longer than their wavelengths, we expect "Ψ -hfl- <->
48 General Description of Wave Propagation in Nonlinear Media The field Ε can generally be decomposed into a longitudinal component E|| parallel to k and a transverse component E± perpendicular to k. The wave equation for E, following C.3), can similarly be sput into two equations: and We now have and JJ- dz C.21a) C.21b) C.22) C-23) Then the approximation of C.20) reduces the second-order differential equa- equation C.21a) to a simple first-order differential equation C-24) This is known as the slowly varying amplitude approximation. The preceding description of the slowly varying amplitude approximation is what is usually found in the literature. However, the real physical implication of the approximation is in neglecting the oppositely propagating field compo- component generated by PNL. Consider, for example, the wave propagation in an isotropic medium, C.25) with plane boundaries at ζ =■ 0 and /. The equation can be solved by the Green function method. Let G(z, z') be the Green function, which obeys the equation C.26) Then we find ilk 'kU-!') forz>z' 'kU l) for ζ < ζ'
Slowly Varying Amplitude Approximation with k = ω-ft/c. The solution of C.25) is given by - Ε— =0 C.28) If we write Ε(ω, ζ) = £F{z)el<k'-"» + iXzJi'«-*1-"" C.29) and impose the boundary conditions d£F/dz' = 0 and d£B/dz' = 0 at ζ = 0" and /+, indicating that no amplitude change should occur outside the medium, then we have = [/,@)«'*+ *,(/)*-*<],-"' C.30) o Comparison of C.28) and C.29) yields and *.(*) = Μ') + i^-f^^'U^^dz'. C.31) The corresponding differential equations for ιίρ and £B are tfz kc2 and ^£ = _I-2H^pNi(fc)i z)e'<*—>. C.32) dz Ac1 Comparing C.24) with C.32), we recognize that C.24) can be obtained by neglecting SB in € (or neglecting $F if 6± is propagating with wavevector - k).
50 General Description of Wave Propagation in Nonlinear Media 3.4 BOUNDARY CONDITIONS The usual boundary conditions for electromagnetic waves should be valid here; for example, the tangential components of Ε and Β at a boundary surface must be continuous for each Fourier component. In general, the solution of the wave equation C.3) for E(k„ ω,) driven by PNL(km, um = ω,) α exp(ikm · r - ίω,ί) has the form where the £H and <fP terms correspond to the homogeneous and particular solutions, respectively. At the boundary, an incoming wave E,(kH,km/, u,} splits into a reflected wave ER(k,R,kmR, ω,) and a transmitted wave Er(k/r, kmT> ω,). Let ζ = 0 be the boundary plane, and k-z be the plane of incidence. Then it is easily seen that the continuity of the tangential compo- components of Ε and Β leads to the following relations:1 and with two similar equations for the y components, and *f/,i = Kml.i ™ k/R.x = ^FFlfi.J = KlT.jt = *«7".jr· C.35) The last equation relating the various tangential wavevectors is most inter- interesting. It prescribes the directions of propagation for all waves (homogeneous and particular) in the media when one of them is given. This is therefore equivalent to Snell's law in linear optics. 3.5 TIME-DEPENDENT WAVE PROPAGATION Propagating waves with time-varying amplitudes should of course obey the time-dependent wave equation in C.1). Here again, the slowly varying ampli- amplitude approximation is usually valid. We expect that both the second-order time derivative and the second-order spatial derivative of the field amplitude can be neglected in the wave equation. This is illustrated in the following by assuming a quasi-monochromatic plane wave propagating along a symmetry axis, z, of the medium. The wave equation takes the form
Time-Dependent Wave Propagation 51 with D(z, i) = EO:, f) + 4irPA)B, f), and Efz, /) = <?<z, r)exp(fc - ίωί). Then, as shown in Section 3.3, the slowly varying amplitude approximation gives —Mz, t) = [ilk-l-*- fcVU«*1-"". C-37) If E(z, i) is expressed in terms of the Fourier integral then we have D(z, f) = fe(u ζ, 0 - ,-2Α^ £ where us = (dk/duy1 is the group velocity. Insertion of C.37) and C.38) in C.36) with the approximation of 5iPNL/3fI 3 -ω:ΡΝ1- yields8 In faci, as we have shown in the time-independent case in Section 3.3, the field amplitude £ in C.39) actually corresponds to SF for the forward propagating wave. For the backward propagating wave, the corresponding equation is Equations C.39) and C.40) should be used for short pulse propagation in a nonlinear medium. The time-derivative term in the equations is negligible only if the amplitude variation is insignificant during the time Τ = ijs/c it takes for light to traverse the medium. We use C.39) and C.40) later in the discussion of nonlinear optical effects with ultrashort pulses.
52 General Description of Wave Propagation in Nonlinear Media REFERENCES 1 See, for example, W. H. Louiseil, Coupled Mode and Parametric Electronics (Wiley, New York, 1960). 1 J. Λ. Armstrong, N. Bloembergen, J. Ducuing, and P. S. Pershan, Phys. Ren. 127, 1918 A962). 3 L Landau and E. M. LifshiE, Electrodynamics in Continuous Media (Addi son-Wesley, Reading, Mass., 1959). ρ 253. A Y. R. Shen, Phys. Reu. 167, 818 A968). 5 P. S. Pershan, Phys. Rev. 130, 919 A963), and many books and review articles on nonlinear optics. 6 See, for example, N. Bloembergen, Nonlinear Optics (Benjamin, New York, 1965). 7 N. Bloembergen and P. S. Pershan, Phys. Rev. 128, 606 A%2). S S. A. Akhmanov, A. S. Chiikin, K. N. Dtabovich, A. I. Kovrigin, R. V. Khokhlov, and A. P. Sukhorukov, IEEEJ. Quant. Electron. QE-4, 598 A94S). BIBLIOGRAPHY Akhmanov, S. Α., and R. V. Khokhlov, Problems of Nonlinear Optics (Gordon and Breath, New York, 1972). Bloembergen, N., Nonlinear Optics (Benjamin, New York, 1%5). Dueuing, in R. Glauber, ed., Quantum Optics, Proceedings of the Inlemalson School of Physics Enrico Fermi Course XUI (Academic Press, New York, 1969), p. 421.
Electrooptical and Magnetooptical Effects Optical properties of a material can be modified by an applied electric or magnetic field. The refractive index changes as functions of the applied electric and magnetic fields are responsible for many electrooptical and magnetooptical effects. Although these effects were well known long before the advent of lasers, they can be regarded as nonlinear optical mixing effects in the limit where one of the field components is of zero or nearly zero frequency. This chapter therefore offers a brief discussion on these effects. 4.1 ELECTROOPTICAL EFFECTS In the presence of an applied dc or low-frequency field E0(Si - 0), the optical dielectric constant ε{ω,Ε0) of a medium is a function of Eo. For sufficiently small Eo, ε(ω,Εο) can be expanded into a power series of Eo: β(ω,Ε0) = ΕA)(ω) + ε'3>(ω + Ο)·Ε0 + ε<3>{" + 20): Ε0Ε0 + ■ ■ ■ . D.1) SinceE + 4ffP = e-E, and Ρ = χA)·Ε + χρ>:ΕΕ + ■■■, we recognize that ε'2>(ω + Ü) = 4πχ™(ω + Ω) and βC)(ω + 2Ω) = 4πχ<3>(ω + 2Ω). D.2) Then, in a medium with no inversion symmetry, the electrooptical effect is dominated by the i<2> term linear in Eo. This is known as Pockel's effect. The symmetry forms of the nonvanishing eB) or χ|2) for (he 20 classes of crystals
54 Ekctrooptical and Magnetooptical Effects are already given in Table 2.1 with, in addition, χ{^(ω = ω + 0) = Χ$\(ω = ω + 0). In a medium with or without inversion symmetry, the quadratic field-dependent term D.1) always exists and is known as the dc Kerr effect. The symmetry forms of eC) or χ|3) for some classes of crystals are given in Table 2.2 with, in addition, χ&,(ω = ω + 0 + 0) = χ%,(ω = ω + 0 + 0) and ν<3! = ν<}> The field-induced refractive indices give rise to linear birefringence or double refraction. Traditionally, the electrooptical effect is defined through the idex ellipsoid1 with ηι;' = (e ^V2 being the refractive index tensor. The power series expan- expansion is carried out for all the coefficients n~.2(E0) of the index ellipsoid The coefficient riJk often is called the linear electrooptical tensor, and pijk/ is the quadratic electrooptical tensor. The values of rijk for many crystals are tabulated in the literature.2 Physically, electrooptical effects result from both ionic or molecular move- movement and distortion of electronic cloud induced by the applied electric field. Even if the induced refractive index change is only around 10"s (typical values of rjjk are around 10~10 to 10"8 cm/v), a medium 1 cm long can already impose on a visible beam a phase retardation of more than ϊγ/2. Therefore, electrooptical effects have been widely used as optical modulators. 4.2 MAGNETOOPTICAL EFFECTS The optical dielectric tensor e of a medium is also a function of an applied dc magnetic field, Ho. It has the symmetry relation 3 ef;(H0)-e,,(-H0). D.5) Here, even in the absence of dissipation, efJ is a complex quantity but it has the property of being Hermitian: eo(H0) = e;,(H0) + «S(H0) - e*(H0). D.6)
Msgnetooptical Effects Then, in a nondissipative medium, we have D 71 Thus the real part of the tensor is symmetric and is an even function of Ho, and the imaginary part is antisymmetric and odd in Ho. The dependence of ejj on Ho leads to circular birefringence or the Faraday effect, while the depen- dependence of ejy on Ho leads to linear birefringence or the Cotton-Mouton effect.3 This can be illustrated with a medium of uniaxial symmetry, having Ho parallel to the axis. In this case, the only nonvanisliing elements of e are t'IX = e'yy and t'ZI even in Ho, and e"y = — t'yx odd in Ho. Diagonalization of e in the coordinate system with orthogonal unit vectors e ± = (x + iy)/^2 and; yields the three diagonal elements e± and ejr, where ε±= t.'xx ± e"y are the susceptibilities for right and left circularly polarized waves, respectively. Since e" « t'IIt the wavevectors of the two circularly polarized waves can be written D.8) and the circular birefringence in a medium of length / is D-9) A linearly polarized beam propagating along ζ will have its polarization rotated by an angle ♦ -ß^fJi D.10) which is known as the Faraday rotation. On the other hand, since t'/x(H0) — e'xx@) is generally different from e'tI(H0) - e'!r@), the linear birefringence in the k-i plane is also altered by the presence of Ho, known as the Cotton-Mouton effect. For sufficiently weak Ho, the power series expansion of e(Hu) yields ε'{ω,Η0) = «·°>(») + β^'ί« + 2Ω)·Η0Η0 + ■ ■ - and Ε"(ωΗ0) = ί"B>(ω + ί2).Η0+ .... D.11)
56 Electrooptical and Magnetooptical Effects Again, eB|/4w, em/A,v, and so on, can be regarded as nonlinear susceptibili- susceptibilities, although they now arise from the magnetic contribution. Analogous to the electrooptical effects, the magnetooptical effects can also be used for optical modulation. The Faraday effect or circular birefringence causes the linear polarization of a beam traversing the medium to rotate. In the low-field limit, the rotation is proportional to the applied magnetic field. Again, the induced change in the dielectric constant or refractive index is usually small (- 10"'/gauss for glass doped with a few percent of rare earth ions), but the rotation resulting form the relative phase shift between the two circular polarizations can be few tens of degrees in a 1-cm-long medium with a field of several thousand gausses. The Cotton-Mouton effect, however, is much weaker and has limited applications. REFERENCES 1 Sec, for example, A. Yariv, Quantum Electronics, 2nd ed. (Wiley, New York, 1975), p. 327. 1 R. J. Pressley, ed., CRC Handbook of Lasers, (Chemical Rubber Co., Cleveland, Obio, 1971), p. 447 3 See, tor example, J. van den Handel, Encyclopedia of Physics, vol. IS, S. Flügge, ed. (Springer- Verlag, Berlin, 1956), p. 15. BIBLIOGRAPHY Landau, L. D,, and Ε. Μ. LifshiE, Electrodynamics in Continuous Media (Pergamon Press, Oxford, 1960). Nye, J. F., Physical Properties of Solids (Ojford University Press, London, 1964). Pressley, R. J., ed.. Handbook of Lasers (Chemical Rubber Co., Cleveland, Ohio, 1971). Wemple, S. H., in F. T. Arecchi and Ε. Ο. Schub-Dubois, eds.. Laser Handbook (North-Holland Publishing Co., Amsterdam, !972), p. 975.
Optical Rectification and Optical Field-Induced Magnetization Modulation and demodulation are commonly known processes at radio wave and microwave frequencies. They form the basis of telecommunications. It is natural to believe that these processes should also exist in the optical region. Light modulation by electrooptical and magnetooptical effects has already been discussed in Chapter 4. In this chapter, optical rectification producing dc electric polarization and magnetization is considered. 5.1 OPTICAL RECTIFICATION In the literature, optical rectification, which was among the first nonlinear optical effects discovered,1 usually refers to the generation of a dc electric polarization by an intense optical beam in a nonlinear medium. The effect can be seen directly from the nonlinear polarization P«),0) = xW(o = ω - ω) :Ε(ω)Ε·(ω) E.1) with Ε(ω) = <fexp(ikT — ί'ωί). The nonlinear susceptibility χ|2ι@ = ω — ω) here governs the magnitude of the effect. In a nonabsorbing medium, the permutation relation of χB) relates χB|@ = ω - ω) to the electrooptical coefficients χ$@ = «-«) = χ%{ω = ω + 0) = χβ>,(« = 0 + ω)
58 Optical Rectification and Optical Field-Induced Magnetuation in a principal-axis system. Thus, from the electrooptical coefficient rjjk, the polarization generated in optical rectification can be predicted. In actual experiments, the induced dc field or voltage, instead of Po>, is measured, bul the two are linearly related. Bass et al.1 and Ward3 performed optical rectifica- rectification experiments and measured *J2i@ = ω ~- ω) for a number of crystals. The experimental arrangement can be simple. A slab of crystal is oriented with its ί axis perpendicular to the two parallel faces of the slab. The faces are coated with silver to form a set of condenser plates. An intense light beam is then directed through (he crystal in a direction perpendicular to ΐ to generate I*-A){0) according to E.1), and the induced dc voltage across the condenser plates is measured. Let the dc dielectric constant of the crystal along / be e0 and assume that the beam intensity can be approximated as uniform over a rectangular cross section s X t in the crystal, as shown in Fig. 5.1. Then, following the infinite plane approximation for condensers, the equations governing the dc fields are Eod~Ea(d-t) + Eht and since there is no net charge on the plates -£„<*.-*) = £„*. E.4) The solution of these equations yields a dc voltage across the plates as E-5) In the experiments to find x'JJ,(O = ω — «), both the voltage V (in the d Ec Τ 1 1< Fig. 5.1 Experimental geometry for measurement of optical rectification.
Effective Free Energy Density 59 mv/MW range) aod the laser intensity must be accurately measured. The results of Bass et al.1 and Ward2 oe optical rectification show thai the identity of E.2) indeed holds within experimental error. S.2 EFFECTIVE FREE ENERGY DENSITY The time-averaged field energy density (U) in a nonlinear medium was derived in Section 3.2, where we saw that we can obtain the polarization P(w)in the medium from the derivative of the effective free energy density if the dispersions of susceptibilities are neglected. Thus the effective free energy density corresponding to a not very intense quasi-monochromatic wave in a nonabsorbing medium subject to a dc electric field Eo is· F[E(w),E0]SF<°>(EQ)-£[xA(E0)£;{u)£/((tJ)+ ...]. E.7a) If χ;1(Ε0) can also be expanded into power series of Eo, it becomes E.7b) t.j.k The free energy density here governs both the electrooptical effect and the optical rectification. From E.6), the polarization induced in optical rectifica- rectification is given by »)£*(«) E-8) as is expected. The same xfj'k in E.7) and E.8) is clearly responsible for the linear electrooptical effect. The above description can be extended to the magnetic case.1·4 For a not too intense light beam propagating in a nonabsorbing medium in the presence of a magnetic field, the effective free energy density can be written as a power •Note that Eo = lifflo^0lI(S2) + EJ(fl}].
60 Optical Rectification and Optical Field-Induced Magnetization series of the optical field E.9) In analogy to the electric case, [χ,,(Η0) — χ^@)] here governs the magneto- opiical effect. Then one also expects from M@) = - 3F/dH0 that there should be a dc magnetization induced in the medium by the incoming light. Indeed, this has been observed and is known as the inverse magnetooptical effect.5 For illustration, we assume a medium of uniaxial symmetry with light propagating along the axis, say, the z-axis. It is well known that if the dc magnetic field is also along z, the two circular polarizations are the eigenmodes of propagation. The effective free energy density can therefore be written as f-F0(H0)-x+(H0)|£+(«)|1-X_(H0)|£_(«)|1+ ■■■ E.10) where χ+= χχχ + ϊχχγ and χ_= χχχ — ϊχχγ are, respectively, the linear sus- susceptibilities for the right and left circularly polarized waves. Equation E.10) can be rearranged into F = F0(H0) - i[X+(ff0) - X.(«o)](|£+i3 " \E-\2) with [χ+(/ϊ0) - x_(//0)l ^Ί [x+t^o) + X-Wo)) being odd and even func- functions of Ho, respectively. The Faraday effect is now proportional to ~3Ü^ + ^ = X+(*o)~x4i/o) ()№ as discussed in Section 4.2. We can also show Xxx{Xo)x„W E13) x«() - x„(o) = *[x+(*o) + x-(ffo) - x+(o) - x-(o)]. This magnetic field-induced susceptibility change is connected to the Cotton-Mouton effect. On the other hand, the optical field-induced magneti- magnetization can also be derived from E.11). With the optical field on, the induced
Inverse Faradaj and Cotlon-Mouron Effects magnetization change along ζ is given by E.14) and The terra &MF even in HQ comes Γτοηι the term responsible for the Faraday effect in F, and iA/CM odd in Ho from the Cotton-Mouton term in F. The phenomena are therefore called the inverse Faraday effect and the inverse Cotton-Mouton effect, respectively. As seen from E.14), even if Ho = 0, the inverse Faraday effect is nonvanishing as long as j£+| * |£_|, and is a maximum for circular polarization. The light beam with |£+|2 Φ t-E-l1 here plays the role of a dc magnetic field and breaks the time-reversal symmetry of the medium. The inverse Cotton-Mouton effect can, however, exist even with linearly polarized light but vanishes for Ηϋ — 0 since d(x++ χ_)/ΘΗ0 is odd in J/o. The effective free energy density then allows us to predict the magnitudes of the inverse Faraday and Cotton-Mouton effects from the measured Faraday rotation and Cotton-Mouton effect in the medium. Physically, the Faraday and Cotton-Mouton effects originate from circular and linear dichroism induced by the dc magnetic field, but how do we describe the inverse effects? This is the subject of the following section. 53 INVERSE FARADAY AND COTTON-MOUTON EFFECTS Microscopically, the light-induced dc magnetization arises because the optical field shifts the different magnetic states of the groiind manifold differently (known as optical Stark shifts), and mixes into these ground states different amounts of excited states. Let the interaction Hamiltonian be From the time-dependent perturbation calculation, we find the perturbed
62 Optical Rectification and Optical Field-Induced Magnetization eigenstate as ■I") - Σ l«')o « + ω' and the optical Stark shift for |n) as E.16) E.Π) where kun.„ = E„. - E„. To demonstrate the inverse magnetooptical effect, we assume here a simple paramagnetic ion with only two pairs of states as shown in Fig. 5.2. The ground | + m) state is connected to the excited | - m') state only by the matrix element { — m'\r_\m) and the 1 — m) state connected to the I + m') state by { + m'|r+j - m), with r±= {x ± iy)/ fl. In an applied magnetic field along ί, the Zeeman splittings for the two pairs are respectively 2gßmH0 and 2g'ßnt'H0, where β is the Bohr magneton. The energy separation between the pairs of states is Αω0 ^> kT al Bo = 0. The dc magnelization of a system of Ν ions per unit volume along ζ is given by •ffgß(Jt) •Ngß[<m\J,\m)f>„ + (- m\J,\- E.18) Here J: is the angular momentum operator and ρ ± m are the thermal popula- Ffe. 5.2 Energy level diagram of an ideal paramagnetic system with only two pans of states connected by circularly polarized opti- cal fields.
Int erst Faraday and Colton-Mouton Effects lions ia\± m) described by the Boltzmann distribution ,-i±n/kT E.19) with E±m = ±gßmH0 + ΔΕ±ηι. Through its perturbation onE±m and \m), the optical field induces a dc magnetization E.20) and - AMP + AMD, ΔΜ" - -Ngßm\{Pm - P_m) -(pi - p°_J], ΔΜ" = -Ngß[(A(m\)j,(A\m))pl +(Δ{ - where p°±m = (ρ±Β,)ΔΕ _0, and ΔΑί'and ΔΛί" come from induced changes in the populations and matrix elements, respectively. From E.15) to E.20) with |AE±J « fcrand|{m'|r+|- m)|2 = |( - m'|r_|m)|2, we readily find Α(ω - ω0) E.21) + g'm')ßH0 (U - ω0J -(gm + g' (gm + g'm')ßH0
64 Optical Rectification and Optical Field-Induced Magnetization and (« - w0) (« + «„)-(«m g'm')ßHa] (« + «ο) +(β» + «'«')№>] i E.22) (« + **)-(gin Since ΔΜΡ arises from the induced population change in the ground states due to optical Stark shifts, ii vanishes in a diamagnetic system which has only a singlet ground state that is populated at ordinary temperatures. It is therefore designated as the paramagnetic part of AM. Because of the finiie relaxation time for the population distribution to reach new equilibrium, &Me cannot respond instantaneously to a short incoming light pulse. In fact, from the time variation of &MF, one should be able to deduce the Tx relaxation time of the ground states. "The &MD term arising from the wavefunciion mixing by the optical field exists even in a diamagnetic system and is designated as the diamagnetic part of ΔΜ. It responds almost instantaneously to the incoming light. The paramagnetic part is proportional to 1/kT for |ΔΕ ± m| ·* W, and the diamagnetic part is essentially independent of temperature. This is similar to the temperature behavior of ordinary paramagnetism and diamagnetism.6 Both &MF in E.21) and ΔΜ° in E.22) have been explicitly decomposed into a
Inverse Faraday and Cotton-Moulon Effects 65 pan proportional to (|£+j2 - \E_\2) and a part proportional to (|£+|2 + 1£_|2). The former corresponds to the inverse Faraday effect, and the latter to the inverse Cotton-Mouton effect. For light frequency far away from reso- resonance, the inverse Cotton-Mouton effect is much smaller than the inverse Faraday effect. As seen from E.21) and E.22), the ratio of iMCM to &MF is about Kg'm' + grri)ßH0/h(u — ωο)| for the paramagnetic part, and is about is'm' + gm)ßHu/h(u ~ «0)|or(p° - p°.m)/(pi + p°-m), whichever is larger, for the diamagnetic part. It may become comparable to 1 when Λ(ω - ω0) approaches the Zeeman splitting energy. However, close to resonance, the induced dc magnetization due to optical pumping often becomes dominant.7 In actual experiments, the inverse Cotton-Mouton effect is distinguishable from the inverse Faraday effect by the fact that with a reversal of the magnetic field Ho, AAiCM changes sign but AMF does not. Finally, we realize that E.21) and E.22) can be derived from E.14) if the microscopic expressions of χ+ and χ_ for the system in Fig. 5.2, following B.17) for χ,7, are used. This is left as an exercise to the readers. The above calculation for AM can of course be generalized to a paramagnetic system with Ν ground states. In dense media, a local field correction factor should also be included. The experimental scheme for observing the light-induced dc magnetization is seen in Fig. 5.3. The light pulse induces a pulsed ΔΛ/(ί) in the sample. The time-derivative ά(ΔΜ)/ώ then induces a voltage across the terminals of a pick-up coil around the sample. As an example, consider the case of CaF2:3%Eu2+. The Faraday rotation at λ = 7000 Ä is 2 χ 10 rad/cm-Oe at 4.2 K. From D.9) and D.10), we obtain [3(χ+- X-)/9H0]Hc_0 = 1.8 χ 100 esu/Oe. Then E.14) predicts that with a circularly polarized ruby laser beam of 10 MW/cm2 in the sample, the induced dc magnetization is ΔΛί = ΔΜ( = 7 x 10 erg/Oe-cm3. yhis is equivalent to that induced by a dc field of about 0.01 Oe. If a ß-switched laser pulse with a 10-MW peak intensity and Laser pulse Polarizer 4 Bean) splitter Oscilloscope Fig. 53 Experimental arrangement for measurement of inverse magnetooptical effects.
66 Optical Rectification and Optical Field-Induced Magnetization a rise time of 2 X 10~8 sec is used, and ΔΛ/(ί) is assumed to follow instanta- instantaneously the intensity variation of the laser pulse, then the voltage induced across the terminals of a 30-tuni pick-up coil wfll be 1.3 mv. This agrees with the experimental observation of van der Ziel et al.,s who demonstrated that the inverse relationship between the Faraday effect and the inverse Faraday effect indeed holds for many paramagnetic and diamagnetic substances. 5.4 INDUCED MAGNETIZATION BY RESONANT EXCITATION The dc magnetization can of course also be induced by light through direct optical pumping, and is generally much stronger than the inverse magnetoopti- cal effect discussed in Section 5.3. Optical pumping by circularly polarized light, for example, alters the population distribution in the magnetic sublevels of both the ground and the excited states. A net angular momentum and hence a magnetization result. Theoretically, rate equations can be used to calculate the population redistribution and hence the magnetization induced by the resonant optical excitation if transition probabilities and relaxation rates between levels are known. Optical pumping in gases and solids has long been a subject of extensive investigation. Polarized fluorescence is often a means for detection of the induced orientation of the angular momentum in the medium. With the setup in Fig. 5.3, however, it can also be studied by measuring the dc magnetization generated in the medium by the laser pulse.7 This may be useful in some cases for studying relaxation between magnetic sublevels in condensed matter. REFERENCES 1 M. Bass, P. A. Franken, J. F. Ward, and G. Weinreich, Phys. Rev. Leu. 9, 446 A962). 2 J. F. Waid, Phys. Rev. 143, 569 A966). 3 Y. R. Shen and N. Bloembergen, Bull. Am. Phys. Soc. 9, 292 A964). 4 P. S. Pershan, I. P. van der Ziel, and L. D. Malmstrom, Phys. Rev. 143, 574 A966). 5 1. P. van der Ziel, P. S. Pershan, and L. D. Malmstrom, Phys. Rev. Uli. 15,190 A965) 6 I. H. van Vleck and M. H. Hebb, Phys. Ren. 46, 17 A934). 7 J. F. Hokrichter, R. M. MacFarlane, and A. L. Schawlow, Phys. Rev. Leu. 26, 652 A971)
Sura-Frequency Generation Wave interaction in a nonlinear medium leads to wave mixing. The result is the generation of waves at sum and difference frequencies. Sum-frequency genera- generation is one of the first three nonlinear optical effects discovered in the early days.1 With the recent advances in tunable lasers, it has become one of the most useful nonlinear optical effects in extending the tunable range to shorter wavelengths. This chapter deals mainly with the basic principle of sum- frequency generation. 6.1 PHYSICAL DESCRIPTION Bass et al.1 in 1962 first observed optical sum-frequency generation in a crystal of triglycine sulfate. In their experiment, two ruby lasers, with their operating wavelengths 10 A apart, were used to provide the input beams. The output, analyzed by a spectrograph, exhibited three lines around 3470 A, two side lines arising from second harmonic generation and the middle one from sum- frequency generation by the two laser beams. The physical interpretation of sum-frequency generation is straightforward. The laser beams at Wj and ω2 interact in a nonlinear crystal and generate a nonlinear polarization Ρρ>(ω3 = wl + ω2)· The latter being a collection of oscillating dipoles acts as a source of radiation at ω3 = ι^ + ω2. In general, the radiation could appear in all directions; the radiation pattern depends on the phase-correlated spatial distribution of PB)(«3). With appropriate arrange- arrangement, however, the radiation pattern can be strongly peaked in a certain direction. This can be determined by phase matching conditions. As discussed in Section 3.1, for effective energy transfer from the pump waves at ωγ and ω2 to the generated waves at «j, in the sum-frequency generation (Fig. 6.1), both energy and momentum conservation must be satisfied. The energy conserva- conservation requires ω^ = ωλ + Wj, while the momentum conservation requires k3 = k! + k2. The latter indicates that the sum-frequency radiation is most effec-
Sum-Frequency Generation Fig. 6.1 Schematic description of sum-frequency generation. tively generated in the so-called phase-matching direction defined by k3 = k, + k2.2 If the wave interaction length / is finite, momentum conservation needs to be satisfied only to within the uncertainty range of l/l. The radiation pattern should therefore be a finite phase-matching peak with an angular width corresponding to Ak - l/l. Absorption in the medium, for instance, can limit the interaction length and broaden the phase-matching peak. In general, sum-frequency generation from the bulk, if allowed and phase-matched, dominates over that from the surface. In reflection, however, because of lack of phase matching, a surface layer ~ λ/2 it thick could contribute significantly to the output. This description gives a qualitative picture of sum-frequency generation, which needs to be substantiated with a more formal treatment. 6.2 FORMULATION The coupled wave approach discussed in Section 3.1 finds a direct application here.3 The three coupled waves are E(wt), E(«2), and the sum-frequency output Ε(ω3). Each field can be decomposed into a longitudinal and a transverse part Ε(ω() = Ε^ω,-) + Ε^ (ω,). They obey the wave equations and V -[Ε,ί«,) + 4πΡ<1>(ωί) + 4πΡ<2>(ω,)] - 0 F.1) where !■(«,) = Ρ,,(ω,) + Ρ± Κ), ΡA)(«ι) - Xffl(«i = "«j + «j) : Ε*(ω2) Ε<ω3), ρΡ>(ωϊ) - χ<2>(«2 = ω3 - «,): E(«,)E·^,), and Ρ^(ω3) = χα\ω3 = W[ + ω2): Ε(ωι)Ε(ωΐ). The general solution of F.1) with boundary condi- conditions is extremely complicated. Fortunately, in real situations, reasonable approximations often can be made to simplify the solution. To illustrate, consider a simple case with the following assumptions: A) all waves are infinite plane waves; B) depletion of energy from the pump waves can be neglected; C) the nonlinear medium is semi-infinite with a plane boundary surface; D) the nonlinear medium is cubic, or the beams are propagating along a symmetry axis. These assumptions are of course not essential, and in the appropriate circumstances can be relaxed, as we shall discuss later.
These assumptions drastically simplify the solution. Negligible depletion of pump field energy means that the nonlinear polarization terms responsible for wave coupling and energy transfer in the equations for E^) and Ε(ω2) can be neglected. Thus the pump waves propagate essentially linearly in the nonlinear medium with Ε(ω,) and Ε(ω2) governed by the linear wave equations. In the infinite plane wave approximation we have in the nonlinear medium Ετ(ωχ) = /lrexp[i(k17. ■ r — ω,ί)] and ΕΓ(ω2) = i2rexp['('tir " r ~ uil)\ Tne only equations left to be solved are those for Ε(ω3) in F.1) with ΡB|(ω3) = ^3B)exp[i(k3j ■ r - w3/)l and k3l = k17- + k2r. The solution for E(w3) in the medium is straightforward. It comprises a homogeneous solution (a free wave with wavevector k3r) and a particular solution (a driven wave with wavevector F.2) where the amplitude A of the homogeneous solution is a coefficient to be determined from the boundary conditions, and we assume ΕΓ||(ω3) + We now give a more complete description of the problem including the boundary conditions.4 Let ζ = 0 be the boundary plane separating the semi- infinite nonlinear medium on the right and a linear medium on the left. All waves are propagating in the x-z plane with wavevectors described in Fig. 6.2. For each ω,, there exists in the linear medium an incoming field Ε,(ω,) from theileft and a reflected field Efi(w;) to the left, and in the nonlinear medium a transmitted field ΕΓ(ω;) to the right. They are related to one another by the boundary conditions. An immediate consequence of matching of the field components at the boundary is that at each ω(, all the wavevector components parallel to the boundary surface must be equal. This leads to Snell's law of reflection and refraction for the pump waves. For the sum-frequency wave, we have In terms of the propagation angles described in Fig. 6.2, this relation becomes F.4) k1Tunö2T Equation F.4) can be regarded as the nonlinear Snell law. When the wavevec-
Fig. 6.2 Description of wavevectors of various waves involved in sum-frequency generation in a semi-inflnite nonlinear medium with a boundary surface al ζ = 0. tors of the incoming pump waves are known, it determines the propagation directions of the nonlinearly generated output waves.4 To complete the solu- solution, one must also find the amplitudes of the output waves. In F.2), #,B>(ω3) = χ*2': SlT€lT. For a given nonlinear medium, χ12) is prescribed, and &1T and £1T are related to the incoming pump field amplitudes by the Fresnel coefficients. The only unknown in the expression of Ej-(w3) in F.2) is the coefficient A. Then we should also consider the incoming and reflected waves at «3, described by Ε,(ω3) and Es(w3), respectively, in the linear medium and F.5) The incoming field amplitude ^3/ is given, but the reflected field amplitude ^3H is to be determined. Thus there are two unknown coefficients, A and SiR, to be fixed by the boundary conditions. Clearly, the requirement that both electric and magnetic field components parallel to the surface must be continuous provides enough relations to solve for A and SiR. We postpone the solution to a later section, considering first the case of sum-frequency generation in the bulk.
Simple Solution of Bulk Sum-Frequency Generation 71 6.3 SIMPLE SOLUTION OF BULK SUM-FREQUENCY GENERATION We are interested here only in sum-frequency generation in the bulk of a nonlinear medium, as described by the growth of ΕΓ(ω3) along ζ in Fig. 6.2. Since the growth of the sum-frequency field is generally insignificant over a distance of a wavelength, the slowly varying amplitude approximation dis- discussed in Section 3.3 is applicable here. With ET(«3) = £iT(z)exp[i(kiT ■ τ - «3 01, F.1) then takes the form dZ W F.6) where Ak = ζ ΔΑ = kir + k27- - k3T F.7) is the phase mismatch. Solution of F.6) yields = 'jr., @) + ir^7^(e'"' " D 6.8) As a further approximation, we may neglect the effect of nonlinear polarisation on reflection and refraction at the boundary. Then, &3T(Q) is directly related to the incoming field *?3/@) through the Fresnel coefficients. The intensity of the generated sum-frequency wave at ζ is given by /,(*)-^p-I^T-i.*)!1. F.9) The corresponding output power is obtained from the integration of I3 over the beam cross section. Here, the finite beam cross section seems to be in conflict with the infinite plane wave assumption, but as is well known, if the beam cross section is sufficiently large, then the ray approximation is valid, and each ray can be treated as an infinite plane wave. Thus with /3 depending on the transverse coordinate p, the total output power at «3 is F.10)
72 Sum-Frequency Generation A case of practical interest occurs in the absence of an input at ω3, that is, ^3/ = 0. In the present approximation, ί3Γ@) also vanishes. Then for |ίτ3/ΔΑ:| » 1, we have |<fJ7.J ·*: |<?37U j,. and the intensity /3 following F.8) and F.9) takes the form 2-αωϊ F.11) As shown in Fig. 6.3, /3 versus Aft;z given here peaks strongly at phase-match- phase-matching ΔΙςζ = 0. The peak value is F.12) and the half-width between the first zeroes is = 2ττ. F.13) Fig. 63 Sum-frequency output as a function of the phase mismatch Ak near &k - 0.
Solution with Boundary Reflection 73 For ζ = 1 cm, kJT ~ I05 cm in a typical example, we find (Δ^)ΐην/λ37- - 10~Λ which indicates that in terms of Ak, the phase-matching peak is often extremely narrow. The calculation of sum-frequency generation in anisotropic media requires slight modification. First, since Pj2* in the second equation of F.1) is usually negligible, ET№(ai)/ETi(ui)= tana, is a constant determined from linear wave propagation, where a, is the angle between Ετ-(ω,) and ΕΓ± (ω,). With the inflnte plane wave approximation, and the slowly varying amplitude approximation, F.1) becomes ζ 6ΊΤ\ζ) 02 k^^ where eJT is the unit vector along SlT. The solution of F.14) is ^-" F15) Within the range of our approximation, F.15) is consistent with F.8) for the isotropic medium. 6.4 SOLUTION WITH BOUNDARY REFLECTION In the more general solution of F.2) and F.5), ΕΓ(ω3) can be rewritten in the form {kL ~ Mr) F.16) with ^3Γ.ι @) = Α + 4*rw^3(i>A2(A:33J - Α32Γ) and *3Γ,|@) - - e,(u3). We then notice immediately that if the approximation is used, F.16) reduces to the simplified solution in F.8). The above approxima- approximation is excellent when ΔΑ is small, or equivalently, when the output in the backward direction with ΔΑ - Ä3rcan be neglected. As pointed out in Section 3.3, the latter is just what the slowly varying amplitude approximation means.
74 Sum-Frequency Generation In finding ^3i-@), however, the more correct solution should include the effect of nonlinear polarization on boundary reflection of the sum-frequency . wave. By requiring the tangential components of electric and magnetic fields be matched at the boundary ζ = 0 (see Fig. 6.4), we find" 1A,2 _ \K3t k.cose. _ ,,2 «3Γ F.18) Fig. 6.4 Schematic diagram describing the incoming and reflected sum-frequency waves in the linear medium and the transmitted sum-frequency waves in the nonlinear medium. The boundary surface between the two media is at ζ — 0-
Solution nilh Boundary Reflection 75 where the subscript k denotes components in the plane of incidence. This set of four equations can then be used to find the four unknown coefficients Ay, Ah, £2Κ y, and (£,„ h. The result is 2(lc2 - k2 \Kls K3T A3Scos ΘΎ X F.19) and Ä:3iicosö3R X (A:37-cosfl3J, - With A and &2R known, the solution in F.2) and F.5) is then complete. It shows that even in the absence of an input, &ir = 0, both &3R and &iT(Q) are nonvanishing because of the nonlinear polarization effect on reflection and refraction. In fact, the reflected sum-frequency wave is easily detectable.5 It can be shown that &3T(Q) = 0 and (?3fl is about kz times smaller than SiT{z) at
76 Sum-Frequency Generation phase matching. Thus the reflected sum-frequency wave is essentially generated by the nonlinear polarization in a surface layer of the order of λ/2 π thick. With some modification, the solution here for a cubic nonlinear medium can be extended to anisotropic media. 6.5 PHASE-MATCHING CONSIDERATIONS As shown in Fig. 6.3, the bulk sum-frequency generation is strong only when \Akz\ < 1. The phase mismatch, Δλ, defines a coherent length, l^ = 1/Ak. If the length of the medium / is below /^, the sum-frequency output increases more or less quadraticaliy with /. If / > /^, the output tends to saturate and may even decrease as / increases. For efficient sum-frequency generation, we must therefore have l^ sufficiently long, of the order of at least a few millimeters in practice. In actual experiments, to avoid reduction of the effective beam interaction length due to finite cross sections, collinear phase matching is required: ΔΑ = kIT + k2T - jfcjj- = 0. F.20) In terms of the refractive indices η(ω(), F.20) can be written as Clearly, for isotropic or cubic materials with normal dispersion /i(w3) > (nfi^), η(ω2)), this relation can never be satisfied. Therefore, collinear phase matching can be achieved only with A) anomalous dispersion or B) birefrin- gent crystals.2 In the latter case, the medium should be a negative uniaxial crystal with «„(«,) < ηο(ω;). By choosing the wave at ui to be extraordinary, it is possible to find [ι((ω3) - n(«[)] and Ιη,(ω3) — ι(ω2)] with opposite signs so that F.21) can be satisfied. Two types of collinear phase matching are commonly used. In Type I, both n(«i) and η(ω:) are ordinary or extraor- extraordinary, while in Type II, either π(ω,) or π(ω,) is ordinary. 6.6 EFFECT OF ABSORPTION Absorption is detrimental to sum-frequency generation since it limits the effective interaction length to roughly the attenuation length. It also broadens the phase-matching peak and lowers the peak value. This can be seen by including absorption in the derivation in Section 6.4. With absorption, the wavevectors become complex: k = k' + iß, where β is the attenuation coeffi- coefficient. Equation F.14) changes into clk'iT jCOS-Oj
Sum-Frequency Genention with High Conversion Efficiency with The resulting solution is c2k;r^cos2a}[iMc' -{ß1T + ß1T - ßiT)\ F 23) If the absorption at either the pump frequencies wL and ω2 or the output frequency ω3 is appreciable so that then the output intensity can be approximated by 2"*ft-«p' F.24) where β = β,Γ + j8;r with &Γ - 0 or β = ftr with β1Γ + β2Γ - 0. The phase-matching curve 13 versus Δλ now takes on a Lorentzian shape with a half-width β. Compared with the zero absorption case, the peak value here is independent of ζ and is reduced by a factor of Χ/β1!1. This shows that with absorption, the effective interaction length is reduced to 1/β, which is just the attenuation length. When both (ßlT+ ß1T) and ßiT are appreciable, the output intensity even decreases exponentially with z. 6.7 SUM-FREQUENCY GENERATION WITH HIGH CONVERSION EFFICIENCY We saw in earlier sections that at perfect phase matching, the output power of sum-frequency generation in a nonabsorbing bulk medium is proportional to I1, the square of the length of the medium. Then as / -* 00, the output power would increase without limit, in violation of energy conservation. This is the consequence of the assumption of negligible pump power depletion, which is not valid when the output becomes significant compared to the pump. The set of three coupled equations in C.4) or F.1) must now be solved together to und a complete solution. For sum-frequency generation with high conversion efficiency, the following conditions usually exist: A) the coupled waves are colhnearly phase matched; B) the medium is nearly lossless; and C) the slowly varying amplitude
78 Sum-Frequency Generation approximation is valid. The coupled equations can therefore be written as [similar to F.14» and with F-25) = ω3 ~ ωι and .. lit From the permutation symmetry of χ'21 in a lossless medium discussed in Section 2.5, we find Kl = Κτ = K} = K. Equation F.25) can be solved exactly.3 First, we can easily show from F.25) that the total power flow W in the medium, 2ιτ [ j 2 3 F.26) is a constant independent of z. This is also known as the Manley-Rowe relation.6 Then the number of photons created at ω3 must be equal to the numbers of photons annihilated at ωι and u2: F.27)
Sum-Frequency Generation wilh High Con*ersion Efficiency In solving F.25), we define j F.28) = κ ml = h|@) + u32@) = u\ + u\ ml ~ Ul(fy + wf @) = Hj + U2 and WI3 = «i@) - «i@) =  ~ u\ Equation F.25) becomes j = UjH2sni0, and -^r = -Hj^sini, F.29) The last equation in F.29) can be integrated to yield ulu2uicos6 = Γ where Γ is a constant independent of z. Then by eliminating sin θ in F.29), we find
80 Sum-Frequency Generation The choice of sign " + " or " — " in F.30) depends on the initial value of Θ. The general solution of F.30) is given in Ref. 3. We consider here the frequently encountered case where the boundary condition is <f3r@) = 0, or w3@) = 0, which leads to Γ = 0 and θ = π/2. Equation F.30) becomes The solution takes the form of a Jacobi elliptical integral dy ■■γι, F.32) ^<Ρ> [(!-/)(!-Τ V)] 1/! assuming m2 = uf@) > ml = ul@) and γ = «2@)/u,@). From F.32) and F.28), we find the intensities of the three waves as and The elliptical function sn^u^OJJ, γ] is periodic in f with a period , _ 2 Γ" Φ Physically, this indicates that as the interaction length increases, energy is transferred back and forth between the wave at u2 and the waves at ω, and ω2 with a period L. While the process first pumps energy into the sum-frequency field, it reverses the energy flow after photons in one of the pump waves are depleted. A simple case of physical interest is the up-conversion process used, for example, to convert an infrared image to the visible. It often occurs with UjfO) » u2@) and ü3@) = 0. Since γ « 1, the elliptical integral of F.32) reduces to a simple form and we find and «?(» = «f@) - ^@)sin2E",(OK] = «i@).
Sum-Frequency Generation with High Conversion Efficiency SI They are plotted in Fig. 6.5, which shows explicitly the periodic variation of energy flow back and forth between the waves at ω2 and ω3. In this case, the depletion of the pump field at ω1 is negligible. Therefore, the solution in F.35) can also be obtained from F.25) by letting &lT be a constant. Another case of interest is when Uj@) = h2@) so that γ = 1. The solution becomes It shows that the period of interaction L is infinite. As- ζ -* co, we have u3(f) -» U[@) and b^J) = n2(f)-* 0. This applies to the case of second harmonic generation, which we discuss in more detail in Chapter 7. The foregoing discussion is based on the assumption of infinite plane waves. In reality, the beam cross sections are finite with intensity variation over the transverse profile. Accordingly, the results here have to be modified, using, for example, the ray approximation. As a result, complete depletion of photons in any beam is impossible. Focused beams are often used in actual experiments to increase the laser intensity, and the theoretical treatment of the problem Fig. 6.5 Relative numbers of photons, as a function of 2, in the three coupled waves with perfect phase matching (ω3 — ι^ + ω2, ft3 - *:, + k2) in an up-conversion pro- process. The initial distribution of the photons in the three waves is Hi - 100 n2 and n3 - 0. (After Ref. 3.)
82 Sum-Frequency Generation becomes more complicated. Boyd and Kleinman7 have worked out the case with negligible depletion. Here we simply refer to their work and postpone our discussion on focused beams to the next chapter. 6.8 A PRACTICAL EXAMPLE In most applications, efficient sum-frequency generation is desired. A number of rules should therefore be followed: 1 A nonlinear optical crystal with little absorption at wl5 ω2, and ω3 is first chosen. It should have a sufficiently large nonlinear susceptibility χ12* and should allow collinear phase matching. 2 The phase matching directions in the crystal, generally in the form of a cone, are determined from the known refractive index tensor of the crystal. 3 The particular phase-matching direction with the appropriate set of polari- polarizations for the three waves is selected to optimize the effective nonlinear susceptibility χ<$ - e3 · χ·2': e$r. 4 The length of the crystal is finally chosen to give the desired conversion efficiency. We consider here a practical example of sum-frequency generation in a KDP crystal with the pump beams at λ, = 5320 Ä and λ2 = 6200 Ä. The sum-frequency generated is at λ3 = 2863 Ä. The ordinary refractive indices of KDP at room temperature are ηο(ωι) = 1-5283, «0(w2) *= 1.5231, and no(w3) = 1.5757. For a beam propagating in a direction at an angle {hJ away from the optical axis, the extraordinary refractive index is given by with «,„(«!) = 1.4822, nem(«2) = 1.4783, and nem(u3) = 1.5231. For type I phase matching, we have from F.21) nf f«j, (if) ) = ^^(«ι) + Y~"o(^2) = 1.5258 from which we can find iff)-sin "l 3)Ί °V " - 3j 76.6°.
Limiting Factors for High Conversion Efficiency 83 Let the waves be propagating in a plane al an angle Φ from the X-axis of the crystal. In the X- Y-Z coordinates, the three polarization vectors are >, -cos$,0) and e3 = ~cqs{HJ cosi>, -cos(/f) sin<£,sin(/f_) I. The KDP crystal has a 4 2-m point group. Its nonvanishing χ12' elements are χΨτζ - X&z = X&y = X%x - 2.6 Χ ΗΓ9 esu and xSn--xft*-2.82xlO-»«u. The effective nonlinear susceptibility* for type I phase matching is Xcff e3 X ■ ei = -2.74 Χ 1 To optimize |χ§||, we must choose Φ = 45°. Finally, in the limit of negligible depletion of pump power, the output power is, following F.12), given by Ρ,Ρ, where ^4 is the beam cross section in square centimeters, ζ in centimeters, and we have used the approximation P, = ltA in megawatts. 6.9 LIMITING FACTORS FOR HIGH CONVERSION EFFICIENCY As a nonlinear effect, the output power of sum-frequency generation is expected to increase with the pump intensity if the pump power is kept Ehe same. This seems to suggest that a tighter focusing of the pump beams should •The expressions of χ'$ for lype I and type II phase matching for the 13 umaxial crystal classes can be found in Ref. 8.
84 Sum-Frequency Generation be used to attain higher conversion efficiency, as long as the longitudinal focal dimension (the confocal parameter) is longer than the effective interaction length. There is, however, a limit to the focusing one can use. First, too high a iaser intensity leads to optical damage in the crystal. Then, the reduced beam cross section due to focusing may decrease the effective interaction length even for collinearly propagating beams. This occurs in an anisotropic crystal. For and extraordinary wave, the directions of wave propagation and ray (energy) propagation are generally different. Therefore, although the waves are propa- propagating collinearly, the rays are not. "Walk-off" of the rays effectively decreases the interaction length. The walk-off effect can of course be minimized if the beams are propagating in a direction along which the waveveclor and ray vector are parallel This may be achieved for sum-frequency generation in a uniaxial crystal in a plane perpendicular to the optical axis, and is known as 90 " phase matching. Such phase matching has been found possible in many crystals over a certain frequency range by temperature tuning. The poor beam quality also reduces the conversion efficiency. A multimode laser beam can be considered crudely as a beam with hot spots. The small dimension of these hot spots increases the walk-off effect and decreases the interaction length. Therefore, for high conversion efficiency, beams with TEM^, mode should be used. Good crystal quality is also important for efficient sum-frequency genera- generation. Inhomogeneity prevents perfect phase matching throughout the crystal. Since {äkz\ < π/2 is needed for efficient energy conversion, the tolerable fluctuation of the refractive index due to inhomogeneity is Δη < λ/4ζ = 2.5 Κ 10~5 for λ = 1 μΐη and ζ = 1 cm. This means that the requirement on the crystal quality is stringent. For the same reason, temperature uniformity throughout the crystal length is also important. For a typical case with dn/dl = 5X 10"s, a temperature stability of ΔΓ< 0.5 Κ throughout the crystal is necessary to achieve Δη < 2.5 X 10"s, This discussion generally applies to all mixing processes. REFERENCES 1 M. Bass, P. A. Franken, Α. Ε Hill, C. W. Peters, and G. Wcinreich, Phys. Rev, Leu. 8, 18 A962). 2 P. D. Maker, R. W. Tirbunc, M. Nisenhoff, and C. Μ Savage, Phys. Rev. Lett, s, 21 A962); J. A. Giordmaine, Phys. Rev, Leu. 8,19 A962). 3 I. A Armstrong, N. Bloemtxrgeo, J. Duelling, and P. S. Pcisaao, Phys. Rev. 127, J918 A962). 4 N. Bloembcrgeo and P. S. Pershan, Phys. Rev. 128, 606 (iM2). 5 J. Ducuing and N. Bloembergen, Phys. Rev. Leu. 10, 474 A963)- R. K. Chang and N. Bloembergen, Phys. Rw. 144, 775 A966). 6 I. M. Man)e> and Η. Ε. Rowe, Proc. IRE 47, 2115 A959).
Bibliography 85 7 G. D. Boyd and D. A. KJeinman, J. Appl. Phys. 39, 3597 A%8). 8 F. Zemike and J. Ε Midwinler, Applied Nonlinear Optics (Wiley, New York, 1973), pp. 64-65. BIBLIOGRAPHY Akhmanov, S. Α., and R. V Khokhlov, Problems of Nonlinear Optics (Gordon and Breach, New York, 1972). Bloembcrgen, N., Nonlinear Optics (Benjamin, New York, 1965). Zernike, F., and J. E. Midwinrer, Applied Nonlinear Optics (Wiley, New York, 1973).
7 Harmonic Generation In the history of nonlinear optics, the discovery of optical harmonic generation marked the birth of the field.1 The effect has since found wide application as a means to extend coherent light sources to shorter wavelengths. This chapter summarizes the important aspects of harmonic generation. As it is a special case of optical mixing, most of the discussion in Chapter 6 can be applied here without much modification. The application of harmonic generation to the measurements of nonlinear optical susceptibilities and to the characterization of ultrashort pulses is also discussed. 7.1 SECOND HARMONIC GENERATION The theory of second harmonic generation follows exactly that of sum-frequency generation discussed in Chapter 7. With ωι = ω2 = ω and ω3 = 2ω, the derivation and results in Sections 6.2 to 6.6 can be applied directly to the present case. In particular, the plane wave approximation with negligible pump depletion yields a second harmonic output power For collinear phase matching in a crystal with normal dispersion, we must have, following F.20), either Μ2ω) = η0(<ο) (type I) G.2) η,Bω) = Η«0(ω) + η£(ω)} (type II). G.3) 8«
Second Harmonic Generation 87 The calculation in the limit of high conversion efficiency requires slight modification. Specifically, for type I collinear phase matching, the permutation symmetiy of χB), following B.30), gives I G-4) ~Τέ« ■ XB)(« = ~« + 2ω): ej2u. The coupled equations of F.24) reduce to the form' 2« _ -ιBω) The conservation of power flow and the conservation of the number of photons in F.26) and F.27), respectively, become 2ω G.6) Using the definition of uu and ulu \μλ and u3 in F.28) with ut = ω2 = ω and ω3 = 2a], we obtain duu ~~jy~ = ~2οωι/2ω5ΐη0 and 1" = u:,sinö G.7) dl If we assume mIu@) = 0, then θ = π/2, and the solution takes the form G.8) The second harmonic output power is then given by Pj-U) = P-@)tantf[c(P.,@)/J4I/1*] G.9)
SS Harmonie Generation where C = £Bω/γι X2wc/i/ε I/2, assuming, for simplicity, k^t = ka = i^2^> = i^Ju.i aad α« = aiw Following G.9), Fig. 7.1 shows how P2w(i) increases with ζ at the expense of Ρω(ζ)- As a practical example, we consider the use of KDP as a second-harmonic generator for a Nd: YAG laser beam at 1.06 μπι. Using the same calculation as in Section 6.8 with ηο(ω) = 1.4939 and ntmBw) = 1.470ό at room tempera- temperature, we find that for type I collinear phase matching, the beams should be propagating in a direction at an angle (β) = 40.5° away from the Z-axis of the crystal. The pump field should be linearly polarized in a plane bisectingjhe X-Z and Y-Z planes in order to yield an optimum χ$ = χψχγBω)$ϊη{Η) = 1.5 X 10"9 esu. With the plane wave approximation, the efficiency of second harmonic generation, following G.9), is I tanh2 4.7 X 1 Ρ @1 G.10) (P.inMW). As seen from G.10), the efficiency η reaches 58% when lPJU)/A]l/2z = 21\/MW, or Ραφ)/Α = 18 MW/crn2 for ζ = 5 cm. In an actual experiment, ij is often less because of the finite dimension and transverse intensity variation of the pump beam. An efficiency as high as 4056 with giant pulses or 85% with ultrashort pulses has, however, been achieved.3 The foregoing calculation does indicate that in order to have an appreciable conversion efficiency (tj > 10%) in a crystal like KDP, a pump intensity of the order of 10 MW/cm2 is needed UI < Q .5 1711VI a. ο η -—β^_^ / -^^^^^ •■^ ^ 1 ^. ■— . 2nd HARMONIC 1 Fig. 7.1 Decay and growth of the normalized fundamental and second harmonic amplitudes, respectively, for the perfect phase-matching case. (After Ref. 1.)
Second Harmonic Generation by Focused Gaussian Beams 89 with a crystal length of a few centimeters. (A much longer crystal is seldom practical.) In general, higher pump intensity leads to a larger η, except in the limit of very high conversion efficiency.1 For a low-power pump beam, therefore, focusing is generally used to increase η and hence the second harmonic output. Focusing, however, increases the walk-off effect, but as mentioned in Section 6.9, it also increases the spread of the beam propagation. The spread hurts the conversion efficiency as a portion of the beam now deviates from the exact phase matching direction. For type I collinear phase matching at the angle \HJ , for example, it can be shown readily that a small deviation Δβ of the beam propagation direction from [H) leads to a phase mismatch3 Ak S iA^gt») -Ύ~- - -^— \m2{h) Δϋ. GΛ1) L»iB») »qB«)J W With Δ/d = irbeing the half-width of the phase-matching peak, the acceptance angle about \HJ one can allow for beam convergence is, from G.11), G12) For a KDP crystal with /= 1 cm and {η) = 45° at λω = 1.06 /im, the acceptance angle ΜΛ is only 2.5 mrad. Equation G.12) shows that Δ0 diverges as [Η) approaches 90°. This occurs because the higher order terms of ΔΟ were neglected in G.11). The correct result for (//_) = 90° is, assuming ηοBω)~ neBw), For a 1-cm KDP crystal at λω = 1.06 /im, the acceptance angle is 36 mrad, which is an order of magnitude larger than in the previous case. The large acceptance angle for 90° phase matching is clearly advantageous if beam focusing is used in second harmonic generation. 7.2 SECOND HARMONIC GENERATION BY FOCUSED GAUSSIAN BEAMS Single-mode laser beams usually are required for efficient second harmonic generation. The conversion efficiency may be greatly enhanced by focusing of the pump beam into the nonlinear crystal. At the focal waist a single-mode
90 Harmonic Generation beam has a Gaussian intensity profile described by expt-p2/!^,2) with Wo being the beam radius. The longitudinal dimension of the focal region is defined by the confocal parameter b = kW£ as the distance between two points on the focusing axis where the beam radii are /Σ times larger than that at the waist. Within the focal region, the beam has approximately a plane wave front, so that the plane wave approximation can be used. We consider first the case of negligible double refraction or walk-off effect, e.g., the 90° phase-matching case. Obviously, if the crystal length / is smaller than the confocal parameter b, the conversion efficiency for second harmonic generation can be described by the result of plane wave approximation in G.10) with A = itWf, Here, as long as b > I, tighter focusing should increase Pu@)/vWq and improve the efficiency. If b < I, however, the approximation breaks down, and tighter focusing tends to reduce the efficiency. Thus optimum focusing occurs when the confocal parameter is about equal to the crystal length, b ~ I. Boyd and Klemman" studied the focusing problem in detail with numerical calcula- calculation. They introduced an efficiency reduction factor A0(i) with £ = l/b to take into account the effect of focusing on the efficiency4'5 They find G.15) The function Ao(£) is plotted in Fig. 7.2. For ξ = l/b < 0.4, we have ha{ ΐ) = ΐ, and ij in G.15) reduces to the form in G.14) as we expected. In the tight focusing limit, ξ > 80, the function Ao(£) lakes the asymptotic form Α0(ξ) = 1.187wVi· ant' lne efficiency drops rapidly with increasing ξ. The maximum value of Α0(ί) = 1.068 appears at f = 2.84, with fto(£) =1 in the range 1 < ί < 6. This shows that although optimum focusing occurs at b = //2.84 for given /, the efficiency will not decrease appreciably even if b — I. With double refraction, the situation is more complicated. Boyd and Klein- man showed that in the limit of low conversion efficiency, G.15) is still valid if A0(i) is replaced by h(B, £) [*<Q. f) - A0(f)l or V = where δ = %p(kj)l/1 is a double refraction parameter, and Ρ = tan S±L
Second Harmonic Generation by Focused Gaussian Beams 10 Fig. 7.2 Efficiency reduction factor h(B, |) versus ξ = i/b at various values of ihe double refraction parameter B. (After Ref. 4.) is the walk-off angle between the Poynting vectors of the collinearly propagat- propagating fundamental and second harmonic waves along a direction at an angle {//) away from the optical axis. The function h( B, £) for several values of Β is seen in Fig. 7.2. Note that h(B, ζ) depends only weakly on ξ near its maximum hM(B). The latter can be approximated to within 10% by the expression3 G.17) with AM@) = 1.068. This equation together with G.16) indicate that the efficiency reduction due to double refraction becomes appreciable when DB2/v)hM@) ~ 1. We can define an effective length Equation G.17) becomes G.18) G.19) This shows explicitly that when 1^ — I in the presence of double refraction, the efficiency for second harmonic generation with optimum focusing reduces by a factor of 2 as compared to the case without double refraction. When lcB -e: /,
Harmonic Generation this efficiency becomes G.20) The effect of double refraction on η,^, is insignificant only when / « 1^. These results do not depend critically on focusing as long as h(B, ξ) = hM(B). One can use a more physical argument to understand these results. Because of double refraction, the phase-matched fundamental and second harmonic beams can overlap only over roughly a distance, /„ = B^vV/p, often known as the aperture length. For optimum focusing, we like to have / = b - kJVj, but to avoid reduction of efficiency by double refraction, we must have / < /„ = \til/kup2, which leads to I < !&= "/Αωρ2, which is the same result described in the preceding paragraph. This discussion assumes that the laser intensity in the crystal is not limited by optical damage. This is of course not always the case. As an example, let us 20 40 60 SO TEMPERATURE 'C Fig. 7-3 Fundamental wavelength versus crystal temperature at 90° phase matching for some of the KDP isomorphs. (After Ref. 6.)
Harmonic Generation in Gases 93 assume a crystal with η = 1-5, / = 1 cm and ρ = 30 mrad at λ_ = 1.06 μτα. Then, Β = 3.65, and from Fig. 7.2, h(B = 3.65, ξ) = hM(B) for 0.2 < £ < 10. Since /eff = 0.04 cm is much less than /, we have ft M (Β) = letc/i, and according to G.20) η cc l^. Comparing with the case of no walk-off and optimum focusing b - /, we have ijp_0 cc /, and hence, vp-o/\ = l/Un = 2ηρ2Ι/λ =* 25. This shows that it is of great advantage to use 90° phase matching to avoid the walk-off effect. The 90° phase matching for second harmonic generation can be achieved by temperature tuning in many crystals. In Fig. 7.3, the 90° phase- matched wavelength as a function of temperature is shown for a number of K.DP isomorphs.6 7.3 THIRD HARMONIC GENERATION IN CRYSTALS In a crystal with inversion symmetry, second harmonic generation is forbidden under the electric dipole approximation, although it can be induced by an applied dc electric field.7 Third harmonic generation, on the other hand, is always allowed. The theory for third harmonic generation in the limit of negligible pump depletion is the same as that for second harmonic generation with P|2)B«) replaced by ΡC)Cω) = χ<3)Cω = ω + α> + ω):Ε(«)Ε(«)Ε(ω). Since |χ{3)| is usually small [- 10" n to ΙΟ"5 esu as compared to|x|2|| - 10"T to 10"* esu typically], and the laser intensity is often limited by optical damage in crystals, the conversion efficiency for this third-order nonlinear process is small. In addition, phase matching is more difficult to achieve. It has therefore found little practical application. An efficient third harmonic generator can, however, be constructed by having two nonlinear crystals in series.8 The first generates a second harmonic beam. The transmitted fundamental beam and (he second harmonic output beam are then combined in the second crystal to yield a third harmonic output by sum-frequency generation. Both processes are phase-matched (either type I or type 11). With a sufficiently intense fundamental beam, the overall efficiency of third harmonic generation can be fairly high. Commercial units with efficiency as high as 20% are available. In principle, this type of two-step third harmonic generation can occur in a single crystal. However, except in very special cases, it is not possible to have both second harmonic generation and sum-frequency generation simulta- simultaneously phase matched. Consequently, the overall conversion efficiency cannot be very significant. 7.4 HARMONIC GENERATION IN GASES Third harmonic generation can also occur in gases. One would think that because of the much lower atomic or molecular density in gases than in liquids or solids, the third-order nonlinear susceptibility |χ|3|| for a gas medium should
Μ Harmonie Generation be much smaller than that for a liquid or solid, and the efficiency for third harmonic generation in gases would be so low that it could never be significant. This conjecture turns out to be incorrect, as pointed out by Miles and Harris.' First, |χC>| can be resonantly enhanced. The much sharper transitions in gases allow much stronger enhancement near resonances, especially those with higher transition matrix elements. Then, the limiting laser intensity in gases is orders of magnitude higher than in condensed matter (> few GW/cnr1 in gases as compared to few hundred MW/cra2 in solids). As a result, although |χC)| is small, the induced nonlinear polarization |PC)| by a high-intensity laser field can be comparable to |PB)| induced in a solid with a moderately intense beam. Consider sodium vapor. The third-order nonlinear susceptibility for Na can be fairly accurately estimated from the general expression of χ|3)Cω) derived by the technique in Sections 2.2 and 2.3: " g.a.b.c where G.21) Here N is the number of atoms per unit volume, and we assume that the frequencies are sufficiently far way from resonances so that the damping constants in the denominators can be neglected. For alkali atoms, the transi- transition frequencies and the major matrix elements are often known. Therefore, |χC>C«)| can be calculated from G.21). This has been done by Miles and Harris.9 The result for Na is seen in Fig. 7.4 along with the energy level diagram for Na. It shows that even when 3ω is a few hundred cm away from an s -* ρ resonance, the near-resonance enhancement can make the value of \χα)[/Ν larger than ΗΓ33 esu. Then, with Ν = W atoms/cm3, and |£(ω)| - 2 Χ 103 esu corresponding to a beam intensity of 1 GW/cm2, the induced nonlinear polarization \Pl>i\ = |χC)£££] can be larger than 10 esu. This is compared with |P|2)B<ö)| = \χα)ΕΕ\ ~ ΚΓ5 esu induced in KDP with |χB'| - 10~9 esu and |£| - 102 esu B.5 MW/cra2). Therefore, third harmonic genera- generation in sodium vapor should be easily observable with 3ω near resonance, e.g., with a Nd laser at 1.06 jim.
Harmonic Generation in Gases 1.064 ft- - .2,6342 Ζ Μ Fig. 7.4 (a) Energy levels of sodium, (fe) Third-order nonlinear polarizabiliiy, X(!)Cii))/tf, versus fundamental wavelength for sodium. (After Ref. 9.) To have high conversion efficiency, aside from resonant enhancement and sufficiently high pump intensity, the third harmonic generation process must be collinearly phase-matched with n(«) = πCω). Since a gas medium is isotropic, the usual method utilizing the birefringent property of a medium for phase matching is not applicable here. Then, phase matching for third harmonic generation (or optical mixing in general) is not always achievable in a gas medium. When anomalous dispersion exists between ω and 3ω, however, it can be achieved by using a buffer gas to compensate the difference of refractive indices at « and 3ω. This is demonstrated in Fig. 7.5. With ω below and 3ω above a strong s — ρ transition of the alkali atom, the anomalous dispersion causes ηΑ(ω) > ηΑ(ΐω) in a pure alkali vapor. If a buffer gas (e.g., Xe) with normal dispersion nfl(u) < nBCw) is mixed into the medium, then by adjust- adjusting the buffer gas density, it is possible to achieve phase matching with Μ") + «β(ω) = "^Cω) + ηΒCω).
Harmonie Generation 5p 6p 5t 4d Μ Sp/ 1.5 1.0 03 0.8 CU INCIDENT WAVELENGTH (μ) <» Fig. 7.4 (Continued). There are several important advantages of using a gas medium for nonlinear optical mixing. 1 A homogeneous medium longer than 10 cm is easily available. 2 Since the medium is isotropic, the walk-off problem does not exist. Opti- Optimum focusing can then be used to increase conversion efficiency. 3 Aside from the high optical damage threshold, a gas medium also has self-healing capability. Except in special cases, no permanent change can be afflicted in the medium by laser-induced ionization or dissociation. 4 Atomic vapor is transparent to radiation at almost all frequencies below the ionization level except for a number of discrete absorption lines, and is the only nonlinear medium one can use in the extreme uv or soft X-ray region. A gas medium may then appear to be ideal for third harmonic generation, especially for conversion to the uv range. High conversion efficiency could presumably be obtained by using a high laser intensity with a reasonably long gas cell. Unfortunately, there are also many factors that often limit the
Harmonic Generation in Gases 0.78 0.35 WAVELENGTH (μ) Ι , INCIDENT WAVELENGTH ( Fig. 7.5 (α) Refractive indices of Rb and Xe versus wavelength. (i>) Required ratio of Xe to alkali atoms versus fundamental wavelength for phase-tnatclied third harmonic generation. (After Ref. 9.) efficiency through limitation of the laser intensity: 1 Linear absorption at ω and 3ω suppresses the efficiency (Section 6.6). Resonant enhancement of |χC)| also enhances the linear absorption, al- although not proportionally. 2 Two-photon and multiphoton absorption may become important in limit- limiting the efficiency when a high-intensity pump beam is used. 3 Population redistribution due to absorption can induce a phase mismatch to the optical mixing process.
98 Harmonie Generation 4 A refractive index change caused by another laser-induced mechanism can also give rise to a phase mismatch. 5 Laser-induced breakdown of the medium may cut off the mixing process. All these factors become rapidly more important when ω or 3ω gets closer to resonance. Usually, C) turns out to be the limiting process and E) may easily occur with long laser pulses. Third harmonic generation in gases has been experimentally demonstrated in many cases.10 With 30-psec, 300-MW Nd:YAG laser pulses optimally focused to a 10-cm2 spot in a 50-cm RbC torr): XeB000 torr) cell, Bloom et al.u observed a phase-matched third harmonic output at 3547 A with a 10% conversion efficiency. The same group also obtained phase-matched third harmonic generation at 3547 Ä in Na: Mg with a 3.8% efficiency. Then, the uv third harmonic generation Irom 5320 to 1773 Ä and from 3547 to 1182 A was also observed in Cd: Ar and Xe: Ar gas mixtures by Kung et al." with a maximum efficiency of 0.3%. This discussion could be extended to higher-order harmonic generation in gases, although the conversion efficiency is expected to be very low because of the relatively small nonlinear optical susceptibilities. It was suggested by Harris13 that coherent vacuum uv and soft X-ray radiation could be obtained from fifth and seventh harmonic generation in atomic vapor. This was later demonstrated by Reintjes et al.14 7.5 MEASUREMENT OF NONLINEAR OPTICAL SUSCEPTIBILITIES The well-established theory of sum-frequency and harmonic generation allows us to deduce nonlinear optical susceptibilities χA)(ω = «L + w2)andxl"l(iiw) from sum-frequency and harmonic generation. We discuss here the measure- measurement of χA)Bω) as an example. As shown in G.1), the absolute magnitude of \elw * χB): ejea\ can be deduced from the measured second harmonic output power if Ρ(ω), A, ifc, z, etc. in G.1) arc known. Then, by choosing the set of polarizations eu and e2w properly aligned with the appropriate crystal orientation, the particular tensor element of χΡ)Bω) can be found. For more accurate determination of χί3)Bω), the second harmonic output PBw) as a function of &kz is measured, and the effect of beam profile is taken into account in the calculation. An absolute measurement is always difficult, however, as the laser beam character- characteristics must be known to great accuracy, ft has only been attempted in a few cases, mainly on ammonium dihydrogen phosphate (ADP).1S The nonlinear susceptibilities of other crystals can then be measured in comparison with that of ADP. In particular, careful comparison between KDP, quartz, and ADP has been made,16 and these three crystals are now often used as reference materials in the measurements of χ·2* of given materials.
Measurement of Nonlinear Optical Susceptibilities Sample BS\ \ PD 1ΡBω) Reference F crystal Fig. 7.6 Experimental arrangement for measurement of the relative second harmonic susceptibility of a sample. In the relative experiment, the laser beam is split into two; one is used for generating Ρ{2ω) in the sample and the other for generating PRBu) in the reference crystal (Fig. 7.6). The ratio of the two is ΡBω) Ρ* B«) 2- · X< 2(\k 1/2) sin2(AkR!R/2) G.22) assuming that two arms have equal laser intensities. Here, the subindex R refers to the reference crystal. With other quantities known, the ratio \elu · Χίϊ>: *ω^υΙ/Ι*ΐω "X5?: ^«*bI can be determined from the measurement of ΡBω)/ΡΛB«) versus Akl. As seen in G.22), the result is now independent of the laser characteristics. This makes the measurement much easier since the very difficult absolute measurement of the laser characteristics is no longer necessary. The result of ΡBω)/ΡΒBω) α sin3(AW/2) as a function of kkl appears as a set of fringes, known as Maker fringes.17 It is usually obtained by rotation of a plane-parallel slab of sample about an axis. The effective sample thickness is then given by dcosff with d being the slab thickness and θ the angle between the slab normal and the beam propagation direction. With ΡBω) α sin2[(AA)i/cos 9/2], the Maker fringes arise through variation of β. In practice, the crystal orientation is also chosen, if possible, to make Δ& independent of 6. An example is seen in Fig. 7.7, where a slab of quartz is used with its c-axis parallel to the face being the axis of rotation for variation of β. The nonlinear susceptibility is in general a complex quantity. The phase factor of x(ifk can be measured from the interference of second harmonic generation in two slabs of crystals in series.18 Let the two crystals of thickness äfj and d2, respectively, be separated by a distance s. Assume phase matching in the first crystal. The second harmonic field generated from the first crystal in
Harmonie Generation t t theory 13.9/i ■40 30 40 20 JO 0 10 ZO ANGULAR ROTATION IN DEGREES Fig. 7.7 Relative second harmonic intensity as a function of the optical thickness of the crystal, displaying the Maker fringes. Change of the optical thickness is achieved by the angular rotation of the crystal. (After Ref, 17.) the normal direction is G.23) The input fields at the entrance of the second crystal are and Ε2Λάι + J) = Elu(dl)e't-2a/c>°'>^^3 G.24) where π 0 is the refractive index of the medium between the two crystals. Then the second harmonic output from the second crystal is PB«) cc 1£ϊω(Α G.25) Expression G.25) shows that ΡBω) depends on the relative phase of the
Measurement ol Nonlinear Optical Susceptibilities 101 effective nonlinear susceptibilities of the two crystals x£'eK and χ^. If the crystals are mounted in an enclosed chamber filled with a known gas and the gas pressure is varied, then because of the dispersion of the gas, ποBω) Φ ηο(ω), the relative phase of the two terms in G.25) will vary, resulting in a set of interference peaks. This observed interference in ΡBω) versus [noBw) — πο(ω)]ί allows us to determine the relative phase of x^>ff and xf?cfr- Usually, XzW of KDP is used as tnc reference. For a nonabsorbing crystal, xf>k is real, either positive or negative. In Table 7.1, we list the values of χ^[ for a number of commonly encountered nonlinear optical crystals. Table 7.1 Selected Second Harmonic Nonlinear Susceptibilities of a Number of Crystals Material o-SiOj (quartz) Te Ba2NaNb50,s LiNbO3 BaTiO, NHeH2PO4 (ADP) KH2PO4 (KDP) ZnO U1O3 CdSe GaAs GaP Symmetry Class 32-O, 32-i>3 mml-C2l, 3m-C3„ 4nvn-C4rl A2m-D2d Ä2m-D2d 6mm-Qa 6-Q 6mm-Qt 43m- Td Um-Td ml 3 ,„·· Υ χ£,= 0.8 + 0.04 χ% - 0.018 X^j = 10* χ™, = -29.1 ± 1.5 χ%- -29.1 + 2.9 xSi - -40 ± 2.9 Xfyy - 6.14 ± 0.56 X&- -11.6 ±1.7 χ® = 81.4 ± 21 χψζχ = -34.4 ±2.8 χ™χ = - 36 + 2.8 χ™ = -13.2 ± 1 χ«^ = 0.96 ± 0.05 X% - 0.97 ± 0.06 χ™ - 0.98 + 0.04 xg = 0.94 xg£ = 4.2 ± 0.4 xlSt = 4.6 ± 0.4 χ®. 14.0 ±0.4 χ*2Λ= -11.2 + 0.6 χ^Ι 8.4 + 2.8 xi2?, = 62 ± 15 X?L = 57 ± 13 χ® = 109 ± 25 χ% - 377 ± 38 X% - 70 Fundamental Wavelength (μηι) 1.0582 10.6 1.0642 1.0582 1.0582 0.6943 1.0582 1.0582 1.0642 10.6 10.6 3.39 "The values of χ*11 are obtained from R. J. Pressley, ed.. Handbook of Lasers (Chemical Rubber Co., Cleveland, Ohio, 1971), p. 497. In the convention we have adopted, our χ121 here are two times larger than the d coefficients given in the literature. Note that χ<2) (esu) here is related to χΙΤι (m/v) by χ (esu) - 3/4* X 10« χB> (m/v).
101 Harmonic Generation For absorbing crystals, x[yi is complex, and Che measurement of second harmonic reflection from the surface, with the help of the theory developed in Section 6.4 is often used to find xijl-19 Again, the interference technique can be adopted to measure the phase of χ{^. The foregoing methods allow accurate relative or absolute measurement of X·^, but the crystal to be studied must be of fair size and good quality. In practice, however, special effort often is needed to grow a crystal of large dimensions. It is therefore important that the nonlinear optical constants of the crystal can somehow be estimated beforehand. The powder method developed by Kurtz20 is most helpful in this respect, Figure 7.S shows the experimental arrangement. Powder sample is packed into a thin cell of a definite thickness, and (he second harmonic output from the sample over the entire 4π solid angle is collected. The output is measured relative to the second harmonic generation in a reference crystal. The desired information can then be obtained by the measurement of the second harmonic output as a function of the panicle size of the powder. For an average panicle size f much smaller than the average coherent length, defused by lcob = w/Älc = itc/«(jiB«) — π(ω)], the second harmonic output ΡBω) increases almost linearly with r since essentially all the particles in the beam are effectively phase matched while the number of particles in the light path decreases inversely wilh f. As f becomes larger than "lail, the output PB«) can increase further if the material is phase-matchable. This is because some particles in the light path should have the correct orientation for phase matching. The output, however, shows saturation as the corresponding decrease in the number of POWDER SAMPLE TRIGGER FOR SCOPE Fig. 7.8 Schematic layout of the apparatus used in the powder measurement of the second-order nonlinearity. [After S. K. Kurtz, IEEE J. Quant. Electron. QE-4, 578 A968).]
Second Harmonic Generation with Ullrashort Pulses Κ I 7/2,-RATIO OF JWG. PAHTICLE SIZE TO ÄVG COHERENCE LENGTH (DIMENSI0NLE3S) Fig. 7.9 Typical second harmonic output as a function of normalized particle size for powders of phase- matchable and non-phase-matchable crystals. [After S. K. Kurtz, IEEE J. Quant. Electron. QE-4, 578 A968).] particles in the light path suppresses the gain of ΡBω) (Fig. 7.9). For non-phase-matchable materials, the output from each particle saturates when f > /„j,, and hence ΡBω) should decrease inversely with rasa result of decrease in the number of particles in the light path as shown in Fig. 7.9. With this technique, numerous materials have been surveyed. They can be divided into five groups:20 centrosymmetric, phase-matchable, non-phase-matchable, large nonlinear coefficient, and small nonlinear coefficient. 7.6 SECOND HARMONIC GENERATION WITH ULTRASHORT PULSES Second harmonic generation with ultrashort pulses requires some special consideration. With the pulse length smaller than the length of the medium, the nonlinear polarization varies drastically along the length at a given time. The only simple case occurs when the group velocities of the forward propagating fundamental and second harmonic waves are the same. Then the two pulses will propagate together and interact with each other as if the problem were stationary. This is the quasi-stationary case. The solution is identical to that of the stationary case if ζ - \>ti replaces ζ where vg is the group velocity. If the group velocity dispersion is nonnegtigible, then the solution becomes much more complicated. Physically, the velocity mismatch causes the fundamental pulse to displace against the second harmonic pulse as they propagate along. This reduces the effective interaction length and decreases the conversion efficiency.21·M A rigorous mathematical treatment of the problem has been worked out by Akhmanov et al.zl Infinite plane waves propagating along ζ with slowly varying amplitudes are assumed. As shown in Section 3.5, the pulse propaga-
104 Harmonie Generation tion of a wave, δ(ζ, t)exp[ikz — iut), in a nonlinear medium can be described by dz d. at kc1 G.26) In the present case, the group velocities of the fundamental and second harmonic waves are vig and u2g, respectively. If we use, as independent variables, ζ and ij s / - z/olg instead of ζ and (, then the wave equations governing the fundamental and second harmonic wave amplitudes &J_z, τι) and £lu(z, η) under the phase-matching condition become -J^ =Ο^ω^2ω- dz G.27) where ν = v2g - o,^ and ο = Birui/klc1)e2a' χ12': &aeu assuming S2u{z = 0) = 0 The solution of G.27) is nontrivial. Akhmanov et al.21 showed that the coupled nonlinear equations can be combined into a single second-order differential equation G.28) where f(ij - vz) = α\&* + Sfj) + fsvd$2Jd-(\ is a function of (η - vz) only, as can be shown by the vanishing Jacobian dF/S-η, dF/dz = 0. Equation G.28) is now a linear equation with a varying coefficient F. It can be solved analytically for arbitrarily large conversion efficiency. Let the fundamental pulse at ζ = 0 take the form G.29) Here, τ is the pulsewidth. In addition, we define a number of characteristic lengths for the problem: L is the length of the medium; L^ = 1/oA is the interaction length at which 75% of the fundamental power is converted into
Second Harmonic Generation with Utrashort Pulses 105 second harmonic power in the stationary case; Lr = τ/ν is the propagating distance over which the overlapping fundamental and second harmonic pulses of width τ are clearly separated. With a new set of variables η = η/τ, ζ a z/L„ τα = vL^h,f= (τ2/τ£ - I)'/2 and £ = [τ2/τ^ - 1]1/2 X jtan"^ — tan(ij - ζ)], the solution of G.27) has the form cosh i + -7SÜ1 G.30) /[l+iü-fI]. When L, » i.NL (τ ~» τα), the group velocity dispersion is clearly negligible so far as second harmonic generation is concerned. In this quasi-stationary case, the solution in G.30) reduces lo They are exactly the same as the stationary solution for second harmonic generation given in G.8) if we replace €u(z = 0) there by A/(l + η2). In the limit of negligible pump depletion, ζ « A + η2)ί-Νί, the second harmonic field from G.31) is proportional to the square of the fundamental field Then the second harmonic pulsewidth is approximately half of that of the fundamental. When Lv < i,NL, the group velocity mismatch becomes important, and the general solution of G.30) must be used. There is a relative displacement between the fundamental and the second harmonic pulses. Consequently, the conversion efficiency drops, and the second harmonic pulse broadens. This has been experimentally confirmed.23 On the other hand, one can also use the group velocity mismatch to sharpen an input second harmonic pulse through amplification. If the fundamental pulse is appreciably longer than the second harmonic pulse, and if og2 > vgl, then the two pulses can be arranged in such a way that the leading edge of the second harmonic pulse always sees the undepleted part of the fundamental pulse and gets amplified more than the lagging edge, resulting in a sharper output pulse.
106 Harmonie Generation The group velocity mismatch is generally more appreciable at higher fre- frequencies because of anomalous dispersion due to absorption bands in the uv region. For a 1-psec pulse propagation in KDP, for example, /,„ s 3 cm at λ = 1.06 μηι and L, = 0.3 cm at λ = 0.53 μΐη. Thus the effect of group velocity mismatch is much more important for frequency doubling of picosec- picosecond pulses into the uv. REFERENCES 1 J. A. Armstrong, N. Bloembergen, J. Ducuing, and P. S. Pershan, Phys. Rev. 127,1918 A962); N. Bloembergen, Nonlinear Optics (Benjamin, New York, 1963), p. 85. 2 J. Rrintjes and R. C. Eckardt, Appl. Phys. Lett. 30,91 A977). 3 R. L. Byer and R. L. Herbst, in Y. R. Shea, ed., Nonlinear Infrared Generation (Springer-Verlag, Berlin, 1977), p. 81. 4 G. D. Boyd and D. A. Kieinmaa, J. Appl. Phys. 39, 3597 A968). 5 D. R. While, E. L. Dawes, and J. H. Marburger, IEEEJ. Quant. Electron. QE-6, 793 A970). 6 K. S. Adhav and R. W. Wallace, IEEE}. Quant. Electron. QE-9, 855 A973). 7 R. W. Terhune, Solid State Design 4, 38 A963). 8 See, for example, the review article by R_ Piston, Laser Focus 14G), 66 A978). 9 R. B. Miles and S, Ε Harris, Appl Phys. Lett. 19, 385 A971); IEEEJ. Quant. Electron. QE-9, 470 A973). 10 J. F. Young, G. C. Bjorklund, A, H. Kung, R. B. Miles, and S. E. Haitis, Phys. Rev. Lett. 17, 1551 A971). 11 D. M. Bloom, G. W. Betters, J. F. Young, and S. E. Harris, Appl. Phys. Leu. 26, 687 A975); D. M. Bloom, J. F. Young, and S. E. Harris, Appl. Phys. Lett. 27, 390 A975). 12 A. H. Kung, I. F. Young, G. C. Bjorklund, and S. Ε Harris, Phys. Rev. Lett, 29, 985 A972); A. H. Kung, J. F. Young, and S. Ε Harris, Appl. Phys. Lett- 22, 301 A973) [Erratum: 18, 239 A976I. 13 S E. Harris, Phys. Rev. Lett. 31, 341 A973). 14 I. Reintjes, C. Y. She, R. C. Eckardt, N. E. Karangelen, R. C. Elton, and R. A. Andrews, Phys. Rev. Lett. 37,1540 A976); Appl. Phys. Lett. 30, 480 A977). 15 G. E. Francois, Phys. Rev. 143, 597 A966); J. E. Bjorkholm and A. B. Siegman, Phys. Rev. 154, 851 A967). 16 J. lerphagnon and S. K. Kurtz, Phys. Rev. Bl, X739 A970). 17 P. D. Mater, R. W. Terhune, M. Nisenoff, and C. M. Savage, Phys. Ree. Lett. 8, 21 A962). IS J. I. Wynne and N. Bluembergen, Phys. Rev, 188, 1211 A969); R. C. Miller and W. A. Nordland, Phys. Heu. B2, 4896 A970). 19 J. Ducuiog and N. Bloembergen, Phys. Rev. Lett. 10, 474 A963); R. K. Chang and N. Bloembergen, Phys. Rev. 144, 775 A966). 20 S. K. Kurtz and T. T. Perry, J. Appl. Phys. 39, 3798 A968); S. K. Kurtz, IEEE J. Quant. Electron. QE-4, 578 A968). 21 S. A. Akhmanov, A. S. Chirkin, K. N. Drabovich, A. I. Kovrigin, R. V. Khokhlov, and A. P. Suthorakov, IEEEJ. Quant. Electron. QE-4, 598 A968). 22 J. Comly and Ε Oannire, Appl. Phys. Lett. 12,7 A968). 23 S. Shapiro, Appl. Phys. Lett. 13,19 A968).
Bibliography BIBLIOGRAPHY Akhmanov, S, Α., A. I. Kovrygin, and A. P. Sukhorukov, in H. Rabin and C. L. Tang, eds., Quantum Electronics (Academic Press, New York, 1972), Vol. 1, ρ 476. Bloembergeu, N., Nonlinear Oplia {Benjamin, New York, 1965). Kleinman, D. Α., in F. T. Arecchi and Ε. Ο. Schulz-Dubois, eds.. Laser Handbook (North-Holland Publishing Co., Amsterdam, 1972), p. 1229. Pressley, R. J., ed., Handbook of Lasers, (Chemical Rubber Co., Cleveland, Ohio, 1971), p. 489. Zernike, F. and J. E. Midwinter, "Applied Nonlinear Optics" (Wiley, New York, 197})
8 Difference-Frequency Generation Difference-frequency generation is theoretically not very different from sum- frequency generation, but the problem is of great technical importance in its own right, as it provides a means for generating intense coherent tunable radiation in the infrared. Traditionally, Wackbody radiation has been the only practical infrared source. Yet, governed by the Planck distribution, it has weak radiative power in the medium- and far-infrared range. A 1-cm2, 5000 Κ blackbody radiates 3500 W over the 4π solid range, but its far-infrared content in the spectral range of 50 ± 1 cm is only 3 Χ 10 W/cm2 ■ sterad. Infrared lasers may seem to have all the desired properties as infrared sources but their output frequencies generally are discrete with essentially zero tunabil- ity. Tunable infrared radiation can, however, be generated by difference- frequency generation. Being coherent with high average or peak intensity, it may find many applications in the field of infrared sciences. This chapter deals mainly with infrared generation by difference-frequency mixing. The diffrac- diffraction effect is considered in the long wavelength limit. Far-infrared generation by ultrashort pulses is also discussed. 8.1 PLANE-WAVE SOLUTION In the infinite plane-wave approximation, the theory for difference-frequency generation follows almost exactly that of sum-frequency generation if the pump intensities can be approximated as constant. Then the output power at ω2 = ω3 ~ ωι generated from the bulk is 8ir3 ,- mi ι ■ ■ ι25ίη2(Δ&ζ/2) , ί3Μ«ιΜ«ιΜ^Τ (Ata/2J (8.1)
Plane-Wave Solution With phase matching and in the presence of appreciable pump depletion, the solution of Section 6.7 must be used. However, the usual initial boundary condition is u2(z = 0) = 0 [i.e., ^@) = 0; the notations here follow those in Section 6.7. ] in the present case. Theequation to be solved becomes(<? = — ir/2) with m, = @) and m3 = h2@). The solution takes the form uj(f) = -Ui(O)sa2[iu3{Q)S, γ], «ϊ(ί) = »?(<>) - w,2@)sn2[;U3@)i, y], (8.3) u^(f) = M^(o) + u12@)sn2[iu3@)i', γ] where dy 1 ' = /-'"i/".(t>> U@),t] -Ό [(i_ In the particular simply case where u^(f) <κ u|@) or |γ2>>2| « 1, we have sn[iw3(O)f] = /sinh|»3@)i'] and hence (8.4) = «Μ») - «?@)siiitf [ For [«3@)ίΊ « 1, this solution leads to (8.1) with ΔΑ - 0. The above results can of course be obtained directly from the coupled wave equations of F.25) by letting g% be a constant. The more general solution with S% = constant, ^@) Φ 0, &ι@) =* 0, and Ak Φ 0 can also be obtained, but this is postponed to the next chapter in connection with a discussion of parametric amplification. The plane-wave approximation adopted here is good as long as the output wavelength is much smaller than the beam cross section. The foregoing results should describe fairly well near-IR and mid-IR generation by difference- frequency mixing. Experimental reports on the subject are numerous, and have been summarized in recent review articles.1·2 An important fact to realize is that the efficiency of infrared generation is expected to be low because of its dependence on the square of the output frequency as seen in (8.1).
110 Difference-Frequency Generation 8.2 FAR-INFRARED GENERATION BY DIFFERENCE-FREQUENCY MIXING The infinite plane-wave approximation ceases to be valid for far-infrared generation in the long-wavelength limil as diffraction becomes important when the pump beam diameter appears comparable to the far-IR wavelength. We must look for a belter solution of the wave equation (8.5) with P(I)(«j) « χ(ϊ>(«2 = ω3 - ω^: Ε(ω3)Ε*(ωι). Since the conversion efficiency is expected to be small because of the small ω2, the depletion of the pump fields can be neglected and the amplitude of Ρσ> can be regarded as independent of propagation. If we neglect the boundary reflections by assuming that the nonlinear crystal is immersed in an infinite index-matched linear medium, the far-field solution of (8.5) has the familiar expression1 where V is the volume of interaction of the pump fields in the nonlinear medium. With P^'fr', ωζ) known, Bfj,u2) can be calculated. Consider, for example, the case where the pump fields E(u,) = ^((r)exp(i7ciz — m^i) and Ε(ω3) = 63(r)exp(ik3z — iu^t) can be approximated by rfj(r) and i3(r) being consianl in a cylinder defined by (x2 + y1) < a2 and vanishing elsewhere. The nonlinear polarization is assumed to have the form A - ft).pw(r', uj) s ppy^'-«*') f0r (*i + /) s „i = 0 for (λ2 + y1) > a1 (8.7) k = It — k With this expression of Pll), the integral in (8.6) can be readily evaluated. Let r = ircos$ + ΧΓΟηφ (see Fig. 8,1), r' = xp'cosff +>>p'sinö + zz', and
Far-Infrared Generation by Difference-Frequency Mixing for the far field. Equation (8.6) then becomes4·5 Ι α with k II Afc \ a =-%-\\ +-r οοβψΙ, β = k7asin4>, ΔΑ: = kls - k2, and Jn being Bessel's function. We now have Integration of c^e(u2)\E(r, (o2)|2/2w over the detector surface (Fig. 8.1) yields the total far-infrared power Ρ(ω2) collected by the detector as (8.10) This result is physically understandable. The [2Jl{ß)/ß\2 term arises from diffraction from a circular aperture as usual, and the (sin a/aI term describes s Lineor M«dle ^ 2o Nonlinter Crysiol OetectoT Fig. 8.1 Schematic for calculations of powet output. The laser beams produce a nonlinear polarization in the crystal at ihe difference frequency; the polarization is then treated as a source for the difference-frequency generation.
(tZ Difference-Frequency Generation the phase-matching condition, (n the limit of k2a 3> 1 so that the diffraction effect is expected to be negligible, 2Jx{ß)/ß is appreciable only for φ s l/fcjU, and then (sin a/aJ reduces approximately to ihe usual phase-matching factor [ sin (AW/2)/(i£//2)p. Also, for k2a s> 1, if the detector is large enough so that φ^^ 3. l/kja, we have /0*m2Ji(ß)/^]Isini» d$ s l/k\a%. The output power Ρ(ω2) calculated from (8.10) can then be shown to have an expression exactly the same as that in (8.1) derived from the infinite plane-wave ap- approximation. The theory here properly takes into account the diffraction effect. Equations (8.9) and (8.10) can in fact be used for an order-of-magnitude estimate of the far-infrared output. In the Jong-wavelength limit, the output approaches toe ω£ dependence on the frequency as one would expect from the dipole radiation theory. This suggests that the efficiency of difference-frequency generation should decrease drastically toward longer wavelengths in the far-IR region. Even so, with the commonly available lasers, the far-IR output from difference-frequency generation still can be much more intense than a blackbody radiation source. A number of simplifying approximations have been employed in the deriva- derivation leading to (8.9). It is possible to use a more realistic expression for Pp>(r', w2) in (8.6) and evaluate the integral numerically to yield a better result. However, the assumption that the nonlinear medium is immersed in an index-matched linear medium is fairly ideal and is usually a poor approxima- approximation. In practice, a nonlinear crystal in air has a very different refractive index at far-IR wavelengths than that of the air. Consequently, reflections of far-IR waves at the boundary surfaces are very important. In treating waves at the boundaries of the nonlinear crystal, one cannot use the /ar-field approxima- approximation. This makes the foregoing theoretical approach inappropriate for dealing with the boundary effects. In order to properly take into account the boundary effects, one should decompose the spatially dependent far-IR Held into spatial Fourier components and impose the boundary conditions on each component separately. The calculation naturally becomes much more complicated, and numerical solution is often necessary for elucidation. We discuss here only some of the physical results and refer the readers to the literature for the details of the calculation.5·6 Since the far infrared refractive index of a solid is usually large (~ 5), reflection at a solid-air boundary can be high. Even multiple reflections can be significant, and in a crystal slab they give rise to a Fabry-Perot factor to each Fourier component ia the output. The long wavelength of the far-IR field makes the phase-matching angle less critical, so that phase matching can be satisfied approximately by far-IR output over a fairly broad cone. This cone is substantially broadened outside the crystal through refraction at the boundary. Part of the far-IR radiation may not even be able to get out of the crystal because of total reflection. Focusing of the pump beams generally helps, but absorption hurts the far-IR output as expected. The output field in (8.6) incorporated with an average transmission coefficient can in fact be a very
Far-Infrared Generation by Ultrasnort Pulses 113 good approximation if the realistic Pffl(r', ω2) is used in the calculation. Equation (8.X) obtained from the infinite plane-wave approximation, however, gives a poor description of the far-IR generation. Experimentally, far-IR generation by difference-frequency mixing has been observed in numerous cases,2·s with output frequencies ranging from 1 to several hundred inverse centimeters. For example, in LiNbO3,x9r-y(w = ωι ~ ω2) = 4.5 X 10~8esu for «! ~ ω2 around the ruby laser frequency. If the pump laser beams are of 1 MW each over an area of 0.2 cm2, a far-IR power of — 3 mW at 10 cm is expected from (8.10) to be generated from a LiNbO3 crystal 0.05 cm thick under the phase-matching condition. In a real experi- experiment, 1 mW at 8.1 cm was detected from a crystal of 0.047 cm.4 Discretely tunable CW far-IR output of 10~7 W has also been observed from mixing of two CO2 lasers B5 W) in GaAs.7 Tunable far-IR radiation can also be generated by stimulated polariton scattering and by spin-flip Raman transi- transitions. We postpone their discussion to Chapter 10. 83 FAR-INFRARED GENERATION BY ULTRASHORT PULSES The discussion in previous sections on infrared generation by optical mixing applies to cases where the pump beams are quasi-monochromatic. The two pump pulses are assumed to be sufficiently long, and the spectral purity of the infrared output, generally correlated with the laser spectral widths, is limited by the pulsewidth. Here, however, we consider the case of far-IR generation by a single short laser pulse.6'9 If the laser pulsewidth is as short as 1 psec, the corresponding spectral width should be at least 15 cm. Then, in a nonlinear crystal, the various spectral components of the pulse can beat with one another and generate far-IR radiation up to the submillimeter range. One might consider this an optical rectification process in which a dc picosecond pulse is generated. However, unlike the case discussed in Section 5.1, we are here interested only in the radiative component of the rectified field. This generation of the radiative output is subject to the influences of phase matching, diffrac- diffraction, boundary reflection, radiation efficiency, and so on.10 Far-IR generation by ultrashort pulses is, as usual, governed by the wave equation (8Λΐ) Given PB)(r, /), (8.11) with appropriate boundary conditions can, at least in principle, be solved. The far-IR output and its power spectrum can then be calculated. The solution of far-IR generation by a single short pulse in a thin slab of nonlinear crystal has actually been obtained through the Fourier
ϊ Ί w ^ S 'S" § - = _ " e <« 5 ε u α ϊ "ο ο « * Ji ** ϊ ^ s us I si »tnaimio ioii5S(is
Far-Infrared Generation by Ultrashort Pulses US transform of E(r, () and PB)(r.'). neglecting the dispersion of ί and χ|2) in PA)(r. I) = X{1> ■ E(r, /)E*(r, r).10 We present here a physical description of the solution. Figure 8.2a shows the calculated power spectrum of far-IR radiation generated by a 2-psec Nd laser pulse from a 1-mm LiNbO, slab. First, the Fabry-Perot geometry of the slab gives rise to the interference pattern under the dotted envelope. Then the dotted envelope of the spectrum is basically the product of three contributions seen in Fig. 8.2(j: curve a represents the power spectrum of the rectified input pulse, curve b describes the ω2 dependence of the radiation efficiency with a much sharper low-frequency cutoff as« ~* 0 due . to diffraction; curve c is the phase-matching curve with its phase-matching peak at ω = 0 for the particular crystal orientation with the c-axis along the face of the slab. Thus, the calculated power spectrum in Fig. 8.2a can be physically understood. Such theoretical calculation actually gives a very good description of the experimental observation. Figure 8.3a shows a comparison between theory and experiment for far-IR generation from a 0.775-mm LiNbO3 with the c-axis in the slab face by a train of normally incident mode-locked Nd/glass laser pulses.8 The Fabry-Perot pattern is absent here because the spectrum has been averaged over the actual instrument resolution. By orienting the crystal to Fig. 83 (aj Far-infrared spectrum generated by mode-locked pulses in LiNbO3 phase matched al zero frequency. Tie experimental points were obtained from the Michelson interferogram and the solid curve from the theoretical calculation assuming Gaussian laser pulses with a 1.8 psec pulsewidth. (b) Far-infrared spectrum generated by mode-locked pulses in LiNbO3 oriented lo have forward and backward phase matching at 13.5 and 6.7 cm, respectively. The experimental points were obtained from the Michelson interferogram. The solid and dashed curves were calculated by assuming Gaussian laser pulses with a pulse-width of 2.3 and 1.8 psec, respectively. (After Ref. A)
116 Difference-Frequency Generation achieve phase matching at finite ω, one expects, from the above discussion, that a single phase-matching peak at ω * 0 may dominate the output spec- spectrum. An example is seen in Fig. 8.3/>. Again, theory and experiment agree well. The two peaks in the theoretical curves correspond to phase-matched generation of the far-IR radiation in the forward and backward directions, respectively. This figure suggests that we can have tunable far-JR output by simply rotating the nonlinear crystal. As shown, the pulse still has a fairly broad linewidth, indicating that it is also a pulse of picosecond duration. Nevertheless, since the output hnewidth is appreciably narrower than the laser linewidth, the output pulse must be appreciably longer than the input pulse. That the output is still significant after the input has more or less decayed away is an interesting fact considering that the medium response to the input pulse is essentially instantaneous in this case. With an input peak power of 0.2 GW over a cross section of ] era2, a far-IR output of 200 W peak power has been detected from a 0.78-mm LiNbO3 crystal.8 REFERENCES 1 R. L. Bycr, in Y. R. Shen, ed. Nonlinear Infrared Generation {Springer-Verlag, Berlin, 1977). 2 F. Zernike, inC. L. Tanked. Methods of Experimental Physics, Vol. XV: Quantum Electronics, Part Β (Academic Press, New York, 1979), p. 143. 3 See, for example, I. D. Jackson, Classical Electrodynamics, 2nd ed. (Wiley, New York, 1975), Chapter 9. 4 D. W. Faries, Κ. Λ. Gehring, P. L. Richards, and Y. R. Shen, Phys. Rev. 180. 363 A969). 5 Y. R Shen, Prog- Quant. Electron. 4, 207 A976). 6 J. R_ Morris and Y. R. Shen, Phys. Rev. A15,1143 A977). 7 B. Lax, R. L. Aggarwal, and G. Favrot, Appl. Phys. Lett. 23, 679 A973); B. Lax and R. L. Aggarwal, J. Opt. Soc. Am. 64, 533 A974). 8 K. H. Yang, P. L. Richards, and Y. R. Shen, Appl. Phys. Lett. 19, 285 A971). 9 T. Yajima and N. Takeuchi, Jap. J. Appl. Phys. 10, 907 A971). 10 J. R. Morris and Y. R. Shen, Optics Comm. 3, 81 A971). BIBLIOGRAPHY Shen, Y. R., Prog. Quant. Electron. 4, 207 A976). Shen, Y. R., ed.. Nonlinear Infrared Generation (Springer-Verlag, Berlin, 1977). Warner, J., in H. Rabin and C. L. Tang, als. Quantum Electronics (Academic Press, New York, 1973), vol. 1, p. 703.
Parametric Amplification and Oscillation The three-wave interaction discussed in previous chapters is manifested by energy flow from the two lower-frequency fields to the sum-frequency field or vice-versa. The latter happens in difference-frequency generation, which, in general, can be initiated with a single pump beam at the sum frequency. Difference-frequency generation can then be considered as the inverse process of sum-frequency generation, and is generally known as a parametric conver- conversion process. Parametric amplification and oscillation in the radio frequency and microwave range were developed before the laser was invented.1 The same process was expected in the optical region, and was actually demonstrated in 1965.2 It has since become an important effect because it allows the construc- construction of widely tunable coherent infrared sources through the controllable decomposition of the pump frequency. In this chapter we explore the theory of parametric fluorescence, amplification, and oscillation together with some practical considerations. 9.1 PARAMETRIC AMPLIFICATION As an inverse process of sum-frequency generation, the general theory of parametric amplification is the same as that for difference-frequency genera- generation. In fact, the only difference of the two processes is in the input conditions. Even there, the difference is not clear-cut, but we normally consider parametric amplification as a process initiated by a single pump beam while difference- frequency generation is initiated by two pump beams of more or less compara- comparable intensities. The difference disappears after a significant fraction of the pump energy has been transferred to the two lower frequency fields. Thus the theoretical description of parametric amplification with infinite plane waves again starts from the set of three coupled wave equations C.4). In the slowly
118 Parametric Amplification and Oscillation varying amplitude approximation with Ε(ω,) = <f,(;)exp[i(k,'r — ω,/ + ψ,)] and a plane boundary at ζ = 0, they become (see Sections 3.3 and 6.7) (9.1) d Tz1 where *-4ί. and hk — and ö0 = φ3 - φ! - φ; is the initial phase difference of the fields at ζ = O.J We assume here % = —τ/2. The solution of (9.1) with Δ£ = 0 has been discussed in previous chapters in connection with sum- and difference-frequency genera- generation. In parametric amplification, E(w3) is known as the pump wave, E(o)j) [or Ε(ω2)] the signal wave, and E(w2) [or E^)] the idler wave. We consider here first the case of negligible pump depletion with Afe Ψ 0. The assumption of negligible pump depletion means that <f3 can be regarded as a constant. Equation (9.1) then reduces to a set of two linearly coupled equations between Sx and $£. Writing St = Cle''"z and S2 — C2e''n', we find immediatdy γλ = ~γ? + Ak, and ~"ί (9.2) This leads to the solution (9.3)
Parametric Amplification 119 This solution shows the following physical properties. If K$s is small so that go < (&kJ, then g is purely imaginary. If K£3 is sufficiently large so that go > (ΔΑ;J, then g is real and positive, and at large gz, both ifj and #2 grow exponential with z. Thus g0 *= (Δ£) is the threshold for parametric amplifica- amplification. The parametric gain is clearly a maximum with g = go at phase matching, Δ/c = 0. Introduction of attenuation coefficients at «1 and ω2 in the above formalism is straightforward. As expected, they increase the threshold and decrease the gain. As an example, consider parametric amplification in LiNbO3 with χ§| = 2.7 X 10~a esu at λ^ - λ2 = 1.06 fim with π} = «2 = 2.23. The maximum gain is found to be g0 = 0.9 X 10" V3 cm. For a pump field of £, = 100 esu corresponding to 2.5 MW/cm2, the gain is 0.9 cm. Thus the overall exponential gain gl even in a crystal of length {= 5 cm is not very large. To achieve an overall gain of g0/ ~ 40, we must either use a pump beam of much higher intensity (which is attainable only with picosecond pulses if optical damage to the crystal is to be avoided) or use an optical cavity to increase the effective length. In the latter case, the system may become an osciJlaior, as wil] be discussed later. As noted in (9.3), the phase mismatch ΔΑ: suppresses the gain very effec- effectively. Therefore, in the limit of high conversion efficiency, we need only consider the phase-matching case although the general solution of (9.1) with ΔΑ Φ 0 has been worked out by Armstrong et al.4 Following the notations and derivation in Section 6.7, we find, assuming θ0 = —n-/2 in (9.1), ±ul-2[ul{ml-ul)(mi-ul)]1/1, (9.4) which has the solution [assuming Wi@) < W;(OI «ι@) Y|,r ιΚ(ο) - «1@) 2 = «ϊ(ο) In the case of u2@) = 0, this result can be shown to reduce to (8.3) derived for difference-frequency generation.